diff --git a/requirements-min.txt b/requirements-min.txt
index b86263de8..d440a6ba1 100644
--- a/requirements-min.txt
+++ b/requirements-min.txt
@@ -2,5 +2,4 @@
 h5py==2.9  # support for setting attrs to lists of utf-8 added in 2.9
 hdmf==2.1.0,<3
 numpy==1.16
-pandas==0.23
 python-dateutil==2.7
diff --git a/requirements.txt b/requirements.txt
index 3669cd3aa..2ad86603f 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,5 +1,4 @@
 h5py==2.10.0
 hdmf==2.1.0
 numpy==1.18.5
-pandas==0.25.3
 python-dateutil==2.8.1
diff --git a/src/pynwb/core.py b/src/pynwb/core.py
index b06edf769..8bd80309e 100644
--- a/src/pynwb/core.py
+++ b/src/pynwb/core.py
@@ -1,6 +1,5 @@
 from h5py import RegionReference
 import numpy as np
-import pandas as pd
 
 from hdmf import Container, Data, DataRegion, get_region_slicer
 from hdmf.container import AbstractContainer, MultiContainerInterface as hdmf_MultiContainerInterface
@@ -222,12 +221,14 @@ def to_dataframe(self):
         '''Produce a pandas DataFrame containing this table's data.
         '''
 
+        import pandas as pd
         data = {colname: self[colname] for ii, colname in enumerate(self.columns)}
         return pd.DataFrame(data)
 
     @classmethod
     @docval(
-        {'name': 'df', 'type': pd.DataFrame, 'doc': 'input data'},
+        # TODO: "real" pd.DataFrame?
+        {'name': 'df', 'type': "pd.DataFrame", 'doc': 'input data'},
         {'name': 'name', 'type': str, 'doc': 'the name of this container', 'default': None},
         {
             'name': 'extra_ok',
@@ -241,6 +242,8 @@ def from_dataframe(cls, **kwargs):
         should match the columns defined on the NWBTable subclass.
         '''
 
+        import pandas as pd
+
         df, name, extra_ok = getargs('df', 'name', 'extra_ok', kwargs)
 
         cls_cols = list([col['name'] for col in getattr(cls, '__columns__')])
diff --git a/src/pynwb/file.py b/src/pynwb/file.py
index fd39b2cc3..c86eba2c3 100644
--- a/src/pynwb/file.py
+++ b/src/pynwb/file.py
@@ -5,7 +5,6 @@
 import copy as _copy
 
 import numpy as np
-import pandas as pd
 
 from hdmf.utils import docval, getargs, call_docval_func, get_docval
 
@@ -669,7 +668,8 @@ def add_stimulus_template(self, timeseries):
         self._add_stimulus_template_internal(timeseries)
         self._update_sweep_table(timeseries)
 
-    @docval({'name': 'data', 'type': (np.ndarray, list, tuple, pd.DataFrame, DynamicTable, NWBContainer, ScratchData),
+    # TODO: bring it pd.DataFrame somehow without causing import of pandas, e.g. if not already loaded?
+    @docval({'name': 'data', 'type': (np.ndarray, list, tuple, DynamicTable, NWBContainer, ScratchData),
              'help': 'the data to add to the scratch space'},
             {'name': 'name', 'type': str,
              'help': 'the name of the data. Only used when passing in numpy.ndarray, list, or tuple',
@@ -683,7 +683,12 @@ def add_stimulus_template(self, timeseries):
     def add_scratch(self, **kwargs):
         '''Add data to the scratch space'''
         data, name, notes = getargs('data', 'name', 'notes', kwargs)
-        if isinstance(data, (np.ndarray, pd.DataFrame, list, tuple)):
+        data_types = [np.ndarray, list, tuple]
+        if 'pandas' in sys.modules:
+            # delayed import and cannot pass DataFrame if pandas is not already loaded
+            import pandas as pd
+            data_types += [ pd.DataFrame ]
+        if isinstance(data, data_types):
             if name is None:
                 raise ValueError('please provide a name for scratch data')
             if isinstance(data, pd.DataFrame):