diff --git a/numba/tests/np/test_ufunc_type_inference.py b/numba/tests/np/test_ufunc_type_inference.py
index 55499973f6d..35288826cf2 100644
--- a/numba/tests/np/test_ufunc_type_inference.py
+++ b/numba/tests/np/test_ufunc_type_inference.py
@@ -51,6 +51,13 @@ def accumulate(ufunc, a):
 def accumulate_dtype(ufunc, a, dtype):
     return numba.typeof(ufunc.accumulate(a, dtype=dtype))
 
+@autojit
+def reduceat(ufunc, a):
+    return numba.typeof(ufunc.reduceat(a))
+
+@autojit
+def reduceat_dtype(ufunc, a, dtype):
+    return numba.typeof(ufunc.reduceat(a, dtype=dtype))
 
 #------------------------------------------------------------------------
 # Tests
@@ -98,7 +105,22 @@ def test_ufunc_accumulate():
 
     equals(accumulate(np.logical_and, a), bool_[:])
 
+    # Test with dtype
+    equals(accumulate_dtype(np.add, a, np.double), double[:])
+
+def test_ufunc_reduceat():
+    equals(reduceat(np.add, a), int32[:])
+    equals(reduceat(np.multiply, a), int32[:])
+
+    equals(reduceat(np.bitwise_and, a), int32[:])
+
+    equals(reduceat(np.logical_and, a), bool_[:])
+
+    # Test with dtype
+    equals(reduceat_dtype(np.add, a, np.double), double[:])
+
 if __name__ == "__main__":
 #    test_binary_ufunc()
 #    test_ufunc_reduce()
     test_ufunc_accumulate()
+    test_ufunc_reduceat()
diff --git a/numba/type_inference/modules/numpymodule.py b/numba/type_inference/modules/numpymodule.py
index 1405c7f6d94..17883704ffe 100644
--- a/numba/type_inference/modules/numpymodule.py
+++ b/numba/type_inference/modules/numpymodule.py
@@ -64,6 +64,12 @@ def promote_to_array(dtype):
         dtype = minitypes.ArrayType(dtype, 0)
     return dtype
 
+def demote_to_scalar(type):
+    "Demote 0d arrays to scalars"
+    if type and type.is_array and type.ndim == 0:
+        return type.dtype
+    return type
+
 def array_from_object(a):
     """
     object -> array type:
diff --git a/numba/type_inference/modules/numpyufuncs.py b/numba/type_inference/modules/numpyufuncs.py
index d5a0fec0edb..6fef4dc5a54 100644
--- a/numba/type_inference/modules/numpyufuncs.py
+++ b/numba/type_inference/modules/numpyufuncs.py
@@ -14,13 +14,28 @@
 from numba.type_inference.modules.numpymodule import (get_dtype,
                                                       array_from_type,
                                                       promote,
-                                                      promote_to_array)
+                                                      promote_to_array,
+                                                      demote_to_scalar)
+
+#----------------------------------------------------------------------------
+# Utilities
+#----------------------------------------------------------------------------
+
+def array_of_dtype(a, dtype, static_dtype, out):
+    if out is not None:
+        return out
+
+    a = array_from_type(a)
+    if not a.is_object:
+        dtype = _dtype(a, dtype, static_dtype)
+        if dtype is not None:
+            return a.copy(dtype=dtype)
 
 def _dtype(a, dtype, static_dtype):
     if static_dtype:
         return static_dtype
     elif dtype:
-        return dtype
+        return dtype.dtype
     elif a.is_array:
         return a.dtype
     elif not a.is_object:
@@ -28,6 +43,10 @@ def _dtype(a, dtype, static_dtype):
     else:
         return None
 
+#----------------------------------------------------------------------------
+# Ufunc type inference
+#----------------------------------------------------------------------------
+
 def binary_map(context, a, b, out):
     if out is not None:
         return out
@@ -67,16 +86,18 @@ def reduce_bool(a, axis, dtype, out):
     return reduce_(a, axis, dtype, out, bool_)
 
 def accumulate(a, axis, dtype, out, static_dtype=None):
-    if out is not None:
-        return out
-
-    dtype = _dtype(a, dtype, static_dtype)
-    if dtype:
-        return promote_to_array(a).copy(dtype=dtype)
+    return demote_to_scalar(array_of_dtype(a, dtype, static_dtype, out))
 
 def accumulate_bool(a, axis, dtype, out):
     return accumulate(a, axis, dtype, out, bool_)
 
+def reduceat(a, indices, axis, dtype, out, static_dtype=None):
+    return accumulate(a, axis, dtype, out, static_dtype)
+
+def reduceat_bool(a, indices, axis, dtype, out):
+    return reduceat(a, indices, axis, dtype, out, bool_)
+
+
 #------------------------------------------------------------------------
 # Binary Ufuncs
 #------------------------------------------------------------------------
@@ -129,8 +150,10 @@ def accumulate_bool(a, axis, dtype, out):
     register_inferer(np, binary_ufunc, binary_map)
     register_unbound(np, binary_ufunc, "reduce", reduce_)
     register_unbound(np, binary_ufunc, "accumulate", accumulate)
+    register_unbound(np, binary_ufunc, "reduceat", reduceat)
 
 for binary_ufunc in binary_ufuncs_compare + binary_ufuncs_logical:
     register_inferer(np, binary_ufunc, binary_map_bool)
     register_unbound(np, binary_ufunc, "reduce", reduce_bool)
     register_unbound(np, binary_ufunc, "accumulate", accumulate_bool)
+    register_unbound(np, binary_ufunc, "reduceat", reduceat_bool)