diff --git a/zap/zap.py b/zap/zap.py
index 363efd5..767e1af 100644
--- a/zap/zap.py
+++ b/zap/zap.py
@@ -81,7 +81,7 @@ def process(cubefits, outcubefits='DATACUBE_ZAP.fits', clean=True,
             zlevel='median', cftype='median', cfwidthSVD=300, cfwidthSP=300,
             nevals=[], extSVD=None, skycubefits=None, mask=None,
             interactive=False, ncpu=None, pca_class=None, n_components=None,
-            overwrite=False, varcurvefits=None):
+            overwrite=False, varcurvefits=None, zero_nan_slices=True):
     """ Performs the entire ZAP sky subtraction algorithm.
 
     This is the main ZAP function. It works on an input FITS file and
@@ -141,7 +141,13 @@ def process(cubefits, outcubefits='DATACUBE_ZAP.fits', clean=True,
         to False.
     varcurvefits : str
         Path for the optional output of the explained variance curves.
-
+    zero_nan_slices : bool
+        In cases where we use a fixed OUTWCS it is very easy that the data 
+        end up having slices with only NaNs. These should be zero'd in the 
+        same way as the AO gaps. If this is set to True, this will be done. It
+        goes through the whole cube - under normal conditions the NaN-only 
+        layers should be at the start and end, but this way we can deal with
+        corner cases.
     """
     logger.info('Running ZAP %s !', __version__)
     t0 = time()
@@ -174,7 +180,8 @@ def _check_file_exists(filename):
         extSVD = SVDoutput(cubefits, clean=clean, zlevel=zlevel,
                            cftype=cftype, cfwidth=cfwidthSVD, mask=mask)
 
-    zobj = Zap(cubefits, pca_class=pca_class, n_components=n_components)
+    zobj = Zap(cubefits, pca_class=pca_class, n_components=n_components,
+               zero_nan_slices=zero_nan_slices)
     zobj._run(clean=clean, zlevel=zlevel, cfwidth=cfwidthSP, cftype=cftype,
               nevals=nevals, extSVD=extSVD)
 
@@ -336,10 +343,12 @@ class Zap(object):
         deconstructed stack
     zlsky : numpy.ndarray
         A 1d array containing the result of the zero level subtraction
-
+    zeroed_nan_slices : list
+        If zero_nan_slices is True, this list records the slices where 
+        NaNs have been set to zero 
     """
 
-    def __init__(self, cubefits, pca_class=None, n_components=None):
+    def __init__(self, cubefits, pca_class=None, n_components=None, zero_nan_slices=True):
         self.cubefits = cubefits
         self.ins_mode = None
 
@@ -383,6 +392,20 @@ def __init__(self, cubefits, pca_class=None, n_components=None):
         else:
             self.notch_limits = None
 
+        # Zero other NaN slices if requested.
+        self.zeroed_nan_slices = []
+        if zero_nan_slices:
+            logger.info('Zeroing slices with only NaNs')
+            nl, ny, nx = self.cube.shape
+            for i in range(nl):
+                # Check whether there are any finite values. If not,
+                # set the whole slice to zero
+                is_finite = np.where(np.isfinite(self.cube[i, :, :]))
+                if len(is_finite[0]) == 0:
+                    self.cube[i, :, :] = 0.0
+                    self.zeroed_nan_slices.append(i)
+
+            
         # NaN Cleaning
         self.run_clean = False
         self.nancube = None
@@ -721,6 +744,9 @@ def make_cube_from_stack(self, stack, with_nans=False):
         if self.ins_mode in NOTCH_FILTER_RANGES:
             lmin, lmax = self.notch_limits
             cube[lmin:lmax + 1] = np.nan
+        if len(self.zeroed_nan_slices) > 0:
+            for i in self.zeroed_nan_slices:
+                cube[i, :, :] = np.nan
         return cube
 
     def remold(self):