diff --git a/LICENSES/PANDAS_PLY_LICENSE b/LICENSES/PANDAS_PLY_LICENSE
new file mode 100644
index 0000000000000..eac7ca890d560
--- /dev/null
+++ b/LICENSES/PANDAS_PLY_LICENSE
@@ -0,0 +1,13 @@
+Copyright 2015 Coursera Inc.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
\ No newline at end of file
diff --git a/pandas/api/tests/test_api.py b/pandas/api/tests/test_api.py
index d4d8b7e4e9747..10a90488020b6 100644
--- a/pandas/api/tests/test_api.py
+++ b/pandas/api/tests/test_api.py
@@ -81,7 +81,7 @@ class TestPDApi(Base, tm.TestCase):
              'pivot', 'pivot_table', 'plot_params', 'qcut',
              'scatter_matrix',
              'show_versions', 'timedelta_range', 'unique',
-             'value_counts', 'wide_to_long']
+             'value_counts', 'wide_to_long', 'X']
 
     # top-level option funcs
     funcs_option = ['reset_option', 'describe_option', 'get_option',
diff --git a/pandas/computation/api.py b/pandas/computation/api.py
index e5814e08c4bbe..a821a579ab6f0 100644
--- a/pandas/computation/api.py
+++ b/pandas/computation/api.py
@@ -2,3 +2,4 @@
 
 from pandas.computation.eval import eval
 from pandas.computation.expr import Expr
+from pandas.computation.delayed import X
diff --git a/pandas/computation/delayed.py b/pandas/computation/delayed.py
new file mode 100644
index 0000000000000..2640f04727d0e
--- /dev/null
+++ b/pandas/computation/delayed.py
@@ -0,0 +1,333 @@
+from __future__ import print_function
+from pandas.compat import iteritems
+from pandas import Series, DataFrame
+
+"""
+Delayed (rename?)
+
+delayed selection api through magic `X` variable
+"""
+
+# includes large portions of pandas_ply, see LICENSES
+
+_error_doc = """
+pandas `X` is a deferred object that cannot be passed into
+most functions. {case} which is invalid.
+To pass a deferred Series into a function, use the .pipe
+function, for example, X.a.pipe(np.log), instead np.log(X.a) """
+
+# numpy / pandas access
+_disallow_attr = [
+    '__array_struct__', '__array_interface__',
+    '_typ', '_data', 'columns', 'values',
+]
+
+_allowed_magic_methods = [
+    '__add__', '__div__', '__sub__', '__truediv__', '__mul__',
+    '__radd__', '__rdiv__', '__rsub__', '__rtruediv__', '__rmul__',
+    '__mod__', '__rmod__',
+    '__eq__', '__ge__', '__gt__', '__lt__', '__le__', '__ne__',
+    '__and__', '__or__', '__invert__' '__neg__', '__pos__',
+    '__rand__', '__ror__',
+    '__abs__', '__pow__',
+]
+# can leave most not implemented, but __iter__ is
+# needed, otherwise __getitem__ sastifies sequence protocol
+# and may iterate forever
+_blacklisted_magic_methods = [
+    '__iter__'
+]
+
+_accessors = [
+    'cat', 'dt', 'str'
+]
+
+
+class Expression(object):
+    """
+    Expression is the (abstract) base class for symbolic expressions.
+    Symbolic expressions are encoded representations of Python expressions,
+    kept on ice until you are ready to evaluate them.
+
+    If an expression is complete, it will act as a 1-argument
+    function, taking a DataFrame whose context to evaluate
+    the expr in.  If not complete, __call__ will create
+    a symbolic call node.
+    """
+
+    def _eval(self, context, **options):
+        """Evaluate a symbolic expression.
+
+        Args:
+            context: The context object for evaluation. Currently, this is a
+                dictionary mapping symbol names to values,
+            `**options`: Options for evaluation. Currently, the only option is
+                `log`, which results in some debug output during evaluation if
+                it is set to `True`.
+
+        Returns:
+            anything
+        """
+        raise NotImplementedError
+
+    def __repr__(self):
+        raise NotImplementedError
+
+    def __getattr__(self, name):
+        """Construct a symbolic representation of `getattr(self, name)`."""
+        if name in _disallow_attr:
+            msg = "The {0} attribuate was called on the object".format(name)
+            raise TypeError(_error_doc.format(case=msg))
+
+        # generally completeness alternates, for instance
+        # in the following expression, marking
+        # incomplete [i], complete [c]
+        #  X . a . pipe(np.exp) . sum()
+        # [i] [c]  [i]    [c]     [i][c]
+
+        # but, accessors may break this pattern, so
+        # as a special case do some introspection
+        # to check if what's being asked for is callable
+        if self._name in _accessors:
+            # cleaner way to do this?
+            acc = getattr(getattr(Series, self._name), name)
+            complete = True
+            if hasattr(acc, '__call__'):
+                complete = False
+        else:
+            complete = not self._complete
+        return GetAttr(self, name, complete)
+
+    def __getitem__(self, name):
+        return GetItem(self, name, complete=True)
+
+    def __call__(self, *args, **kwargs):
+        if self._complete:
+            # selection lambda passed to pandas
+            if len(args) != 1:
+                msg = ("too many values passed into `X`, selection "
+                       "likely malformed")
+                raise ValueError(msg)
+            df, = args
+            if not isinstance(df, DataFrame):
+                msg = ("`X` selection can only be evaluated in the context "
+                       "of a DataFrame ")
+                raise ValueError(msg)
+            return self._eval({0: df})
+        # symbolic call
+        return Call(self, args=args, kwargs=kwargs)
+
+    # error trapping
+    def __array__(self, *args, **kwargs):
+        msg = "The object was attempted to be converted to a numpy array"
+        raise TypeError(_error_doc.format(case=msg))
+
+
+def _get_sym_magic_method(name):
+    def magic_method(self, *args, **kwargs):
+        return Call(GetAttr(self, name), args, kwargs)
+    return magic_method
+
+
+def _get_blacklisted_method(name):
+    def blacklisted_method(self, *args, **kwargs):
+        msg = "The {0} method was called".format(name)
+        raise TypeError(_error_doc.format(case=msg))
+    return blacklisted_method
+
+for name in _allowed_magic_methods:
+    setattr(Expression, name, _get_sym_magic_method(name))
+for name in _blacklisted_magic_methods:
+    setattr(Expression, name, _get_blacklisted_method(name))
+
+
+class Symbol(Expression):
+    """`Symbol(name)` is an atomic symbolic expression, labelled with an
+    arbitrary `name`."""
+
+    def __init__(self, name, complete=False):
+        self._name = name
+        self._complete = complete
+
+    def _eval(self, context, **options):
+        if options.get('log'):
+            print('Symbol._eval', repr(self))
+        result = context[self._name]
+        if options.get('log'):
+            print('Returning', repr(self), '=>', repr(result))
+        return result
+
+    def __repr__(self):
+        return 'Symbol(%s)' % repr(self._name)
+
+
+class GetAttr(Expression):
+    """`GetItem(obj, name)` is a symbolic expression representing the result of
+    `getattr(obj, name)`. (`obj` and `name` can themselves be symbolic.)"""
+
+    def __init__(self, obj, name, complete=True):
+        self._obj = obj
+        self._name = name
+        self._complete = complete
+
+    def _eval(self, context, **options):
+        if options.get('log'):
+            print('GetAttr._eval', repr(self))
+        evaled_obj = eval_if_symbolic(self._obj, context, **options)
+        result = getattr(evaled_obj, self._name)
+        if options.get('log'):
+            print('Returning', repr(self), '=>', repr(result))
+        return result
+
+    def __repr__(self):
+        return 'getattr(%s, %s)' % (repr(self._obj), repr(self._name))
+
+
+class GetItem(Expression):
+    """`GetAttr(obj, name)` is a symbolic expression representing the result of
+    `getattr(obj, name)`. (`obj` and `name` can themselves be symbolic.)"""
+
+    def __init__(self, obj, name, complete=True):
+        self._obj = obj
+        self._name = name
+        self._complete = complete
+
+    def _eval(self, context, **options):
+        if options.get('log'):
+            print('GetItem._eval', repr(self))
+        evaled_obj = eval_if_symbolic(self._obj, context, **options)
+        result = evaled_obj[self._name]
+        if options.get('log'):
+            print('Returning', repr(self), '=>', repr(result))
+        return result
+
+    def __repr__(self):
+        return 'getitem(%s, %s)' % (repr(self._obj), repr(self._name))
+
+
+class Call(Expression):
+    """`Call(func, args, kwargs)` is a symbolic expression representing the
+    result of `func(*args, **kwargs)`. (`func`, each member of the `args`
+    iterable, and each value in the `kwargs` dictionary can themselves be
+    symbolic)."""
+
+    def __init__(self, func, args=None, kwargs=None, complete=True):
+        self._func = func
+        if not args:
+            args = []
+        if not kwargs:
+            kwargs = {}
+        self._args = args
+        self._kwargs = kwargs
+        self._complete = True
+        self._name = None
+
+    def _eval(self, context, **options):
+        if options.get('log'):
+            print('Call._eval', repr(self))
+        evaled_func = eval_if_symbolic(self._func, context, **options)
+        evaled_args = [eval_if_symbolic(v, context, **options)
+                       for v in self._args]
+        evaled_kwargs = dict((k, eval_if_symbolic(v, context, **options))
+                             for k, v in iteritems(self._kwargs))
+        result = evaled_func(*evaled_args, **evaled_kwargs)
+        if options.get('log'):
+            print('Returning', repr(self), '=>', repr(result))
+        return result
+
+    def __repr__(self):
+        return '{func}(*{args}, **{kwargs})'.format(
+            func=repr(self._func),
+            args=repr(self._args),
+            kwargs=repr(self._kwargs))
+
+
+def eval_if_symbolic(obj, context, **options):
+    """Evaluate an object if it is a symbolic expression, or otherwise just
+    returns it back.
+
+    Args:
+        obj: Either a symbolic expression, or anything else (in which case this
+            is a noop).
+        context: Passed as an argument to `obj._eval` if `obj` is symbolic.
+        `**options`: Passed as arguments to `obj._eval` if `obj` is symbolic.
+
+    Returns:
+        anything
+
+    Examples:
+        >>> eval_if_symbolic(Symbol('x'), {'x': 10})
+        10
+        >>> eval_if_symbolic(7, {'x': 10})
+        7
+    """
+    return obj._eval(context, **options) if hasattr(obj, '_eval') else obj
+
+
+def to_callable(obj):
+    """Turn an object into a callable.
+
+    Args:
+        obj: This can be
+
+            * **a symbolic expression**, in which case the output callable
+              evaluates the expression with symbols taking values from the
+              callable's arguments (listed arguments named according to their
+              numerical index, keyword arguments named according to their
+              string keys),
+            * **a callable**, in which case the output callable is just the
+              input object, or
+            * **anything else**, in which case the output callable is a
+              constant function which always returns the input object.
+
+    Returns:
+        callable
+
+    Examples:
+        >>> to_callable(Symbol(0) + Symbol('x'))(3, x=4)
+        7
+        >>> to_callable(lambda x: x + 1)(10)
+        11
+        >>> to_callable(12)(3, x=4)
+        12
+    """
+    if hasattr(obj, '_eval'):
+        return lambda *args, **kwargs: obj._eval(
+            dict(enumerate(args), **kwargs))
+    elif callable(obj):
+        return obj
+    else:
+        return lambda *args, **kwargs: obj
+
+
+# keep?
+def sym_call(func, *args, **kwargs):
+    """Construct a symbolic representation of `func(*args, **kwargs)`.
+
+    This is necessary because `func(symbolic)` will not (ordinarily) know to
+    construct a symbolic expression when it receives the symbolic
+    expression `symbolic` as a parameter (if `func` is not itself symbolic).
+    So instead, we write `sym_call(func, symbolic)`.
+
+    Tip: If the main argument of the function is a (symbolic) DataFrame, then
+    pandas' `pipe` method takes care of this problem without `sym_call`. For
+    instance, while `np.sqrt(X)` won't work, `X.pipe(np.sqrt)` will.
+
+    Args:
+      func: Function to call on evaluation (can be symbolic).
+      `*args`: Arguments to provide to `func` on evaluation (can be symbolic).
+      `**kwargs`: Keyword arguments to provide to `func` on evaluation (can be
+          symbolic).
+
+    Returns:
+        `ply.symbolic.Expression`
+
+    Example:
+        >>> sym_call(math.sqrt, Symbol('x'))._eval({'x': 16})
+        4
+    """
+
+    return Call(func, args=args, kwargs=kwargs)
+
+X = Symbol(0)
+"""A Symbol for "the first argument" (for convenience)."""
diff --git a/pandas/computation/tests/test_delayed.py b/pandas/computation/tests/test_delayed.py
new file mode 100644
index 0000000000000..54c6d2bbddd10
--- /dev/null
+++ b/pandas/computation/tests/test_delayed.py
@@ -0,0 +1,21 @@
+import pandas as pd
+from pandas import X
+
+import pandas.util.testing as tm
+from pandas.util.testing import assert_frame_equal
+
+
+class TestDelayedApi(tm.TestCase):
+    def test_basics(self):
+        df = pd.DataFrame({'a': [1, 2, 3], 'b': [1.5, 2.5, 3.4],
+                           'c': ['abc', 'def', 'efg'],
+                           'd': pd.to_datetime(['2014-01-01',
+                                                '2014-01-02', '2014-01-03'])})
+        assert_frame_equal(df[df['a'] > 1], df[X.a > 1])
+        assert_frame_equal(df[df['a'] == 1], df[X.a == 1])
+        assert_frame_equal(df.assign(e=lambda x: x['b'] + 1),
+                           df.assign(e=X.b + 1))
+        assert_frame_equal(df.assign(e=lambda x: x['d'].dt.day),
+                           df.assign(e=X.d.dt.day))
+        assert_frame_equal(df.assign(e=lambda x: x['c'].str.upper()),
+                           df.assign(e=X.c.str.upper()))