diff --git a/convoys/utils.py b/convoys/utils.py
index e903623..898a5b2 100644
--- a/convoys/utils.py
+++ b/convoys/utils.py
@@ -1,4 +1,5 @@
 import datetime
+import numpy
 import pandas
 
 __all__ = ['get_arrays']
@@ -78,8 +79,8 @@ def get_arrays(data, features=None, groups=None, created=None,
         G = data[groups].apply(lambda g: group2j.get(g, -1)).values
         res.append(G)
     else:
-        groups_list = []
-        X = data[features].values
+        groups_list = None
+        X = numpy.array([numpy.array(z) for z in data[features].values])
         res.append(X)
 
     # Next, construct the `B` and `T` arrays
diff --git a/test_convoys.py b/test_convoys.py
index d90be51..bba8d2f 100644
--- a/test_convoys.py
+++ b/test_convoys.py
@@ -241,6 +241,14 @@ def test_convert_dataframe():
     # TODO: assert things
 
 
+def test_convert_dataframe_features(n=1000):
+    df = _generate_dataframe(n=n)
+    df['features'] = [tuple(numpy.random.randn() for z in range(3)) for g in df['group']]
+    df = df.drop('group', axis=1)
+    unit, groups, (X, B, T) = convoys.utils.get_arrays(df)
+    assert X.shape == (n, 3)
+
+
 def _test_plot_cohorts(model='weibull', extra_model=None):
     df = _generate_dataframe()
     unit, groups, (G, B, T) = convoys.utils.get_arrays(df)