diff --git a/xray/backends/api.py b/xray/backends/api.py
index afc85a4d716..b4084bb71ff 100644
--- a/xray/backends/api.py
+++ b/xray/backends/api.py
@@ -39,7 +39,7 @@ def _get_default_engine(path, allow_remote=False):
 def open_dataset(filename_or_obj, group=None, decode_cf=True,
                  mask_and_scale=True, decode_times=True,
                  concat_characters=True, decode_coords=True, engine=None,
-                 chunks=None, lock=True):
+                 chunks=None, lock=True, close_files=False):
     """Load and decode a dataset from a file or file-like object.
 
     Parameters
@@ -84,6 +84,9 @@ def open_dataset(filename_or_obj, group=None, decode_cf=True,
         If chunks is provided, this argument is passed on to
         :py:func:`dask.array.from_array`. By default, a lock is used to avoid
         issues with concurrent access with dask's multithreaded backend.
+    close_files: bool, optional
+        If True, the engine should try to keep the underlying files closed.
+        Currently only supported for engine='scipy'.
 
     Returns
     -------
@@ -108,6 +111,9 @@ def maybe_decode_store(store):
             ds = ds.chunk(chunks, lock=lock)
         return ds
 
+    if close_files and engine != 'scipy':
+        raise ValueError('close_files only supported for scipy engine')
+
     if isinstance(filename_or_obj, backends.AbstractDataStore):
         store = filename_or_obj
     elif isinstance(filename_or_obj, basestring):
@@ -137,7 +143,7 @@ def maybe_decode_store(store):
             if engine == 'netcdf4':
                 store = backends.NetCDF4DataStore(filename_or_obj, group=group)
             elif engine == 'scipy':
-                store = backends.ScipyDataStore(filename_or_obj)
+                store = backends.ScipyDataStore(filename_or_obj, close_files=close_files)
             elif engine == 'pydap':
                 store = backends.PydapDataStore(filename_or_obj)
             elif engine == 'h5netcdf':
@@ -147,7 +153,10 @@ def maybe_decode_store(store):
                                  % engine)
 
         with close_on_error(store):
-            return maybe_decode_store(store)
+            ds = maybe_decode_store(store)
+            if close_files:
+                store.ds.close()
+            return ds
     else:
         if engine is not None and engine != 'scipy':
             raise ValueError('can only read file-like objects with '
diff --git a/xray/backends/scipy_.py b/xray/backends/scipy_.py
index 872c55fe2ce..7eb801671a2 100644
--- a/xray/backends/scipy_.py
+++ b/xray/backends/scipy_.py
@@ -3,6 +3,8 @@
 import numpy as np
 import warnings
 
+import scipy.io
+
 from .. import Variable
 from ..conventions import cf_encoder
 from ..core.pycompat import iteritems, basestring, unicode_type, OrderedDict
@@ -28,6 +30,33 @@ def _decode_attrs(d):
                        for (k, v) in iteritems(d))
 
 
+class ScipyNetcdfWrapper(scipy.io.netcdf_file):
+    # wrap scipy.io.netcdf_file object in a class that
+    # can open or close the underlying file as needed
+    def __init__(self, *args, **kwargs):
+        self._is_open = True
+        self._variables = dict()
+        self._init_args = args
+        self._init_kwargs = kwargs.copy()
+        self.close_files = kwargs.pop('close_files')
+        super(ScipyNetcdfWrapper, self).__init__(*args, **kwargs)
+
+    def close(self):
+        super(ScipyNetcdfWrapper, self).close()
+        self._is_open = False
+
+    @property
+    def variables(self):
+        # check to see if the underlying file is closed
+        if not self._is_open:
+            # if so, reinitialize it
+            self.__init__(*self._init_args, **self._init_kwargs)
+        return self._variables
+
+    @variables.setter
+    def variables(self, value):
+        self._variables = value
+
 class ScipyArrayWrapper(NumpyIndexingAdapter):
     def __init__(self, netcdf_file, variable_name):
         self.netcdf_file = netcdf_file
@@ -39,7 +68,10 @@ def array(self):
         # because otherwise scipy complains about variables or files still
         # referencing mmapped arrays when we try to close datasets without
         # having read all data in the file.
-        return self.netcdf_file.variables[self.variable_name].data
+        a = self.netcdf_file.variables[self.variable_name].data
+        if self.netcdf_file.close_files:
+            self.netcdf_file.close()
+        return a
 
     @property
     def dtype(self):
@@ -53,6 +85,8 @@ def __getitem__(self, key):
         # after closing associated files.
         copy = self.netcdf_file.use_mmap
         data = np.array(data, dtype=self.dtype, copy=copy)
+        if self.netcdf_file.close_files:
+            self.netcdf_file.close()
         return data
 
 
@@ -65,9 +99,8 @@ class ScipyDataStore(AbstractWritableDataStore):
     It only supports the NetCDF3 file-format.
     """
     def __init__(self, filename_or_obj, mode='r', format=None, group=None,
-                 writer=None, mmap=None):
+                 writer=None, mmap=None, close_files=False):
         import scipy
-        import scipy.io
         if mode != 'r' and scipy.__version__ < '0.13':  # pragma: no cover
             warnings.warn('scipy %s detected; '
                           'the minimal recommended version is 0.13. '
@@ -87,16 +120,25 @@ def __init__(self, filename_or_obj, mode='r', format=None, group=None,
             raise ValueError('invalid format for scipy.io.netcdf backend: %r'
                              % format)
 
+        # can't use mmap with close_files
+        if close_files:
+            mmap = False
+
         # if filename is a NetCDF3 bytestring we store it in a StringIO
         if (isinstance(filename_or_obj, basestring)
                 and filename_or_obj.startswith('CDF')):
             # TODO: this check has the unfortunate side-effect that
             # paths to files cannot start with 'CDF'.
             filename_or_obj = BytesIO(filename_or_obj)
-        self.ds = scipy.io.netcdf_file(
-            filename_or_obj, mode=mode, mmap=mmap, version=version)
+        self.ds = ScipyNetcdfWrapper(
+            filename_or_obj, mode=mode, mmap=mmap, version=version,
+            close_files=close_files)
         super(ScipyDataStore, self).__init__(writer)
 
+        if close_files:
+            self.ds.close()
+        
+
     def store(self, variables, attributes):
         # All Scipy objects get CF encoded by default, without this attempting
         # to write times, for example, would fail.