diff --git a/doc/source/whatsnew/v1.0.2.rst b/doc/source/whatsnew/v1.0.2.rst
index 1b6098e6b6ac1..fae41953a6371 100644
--- a/doc/source/whatsnew/v1.0.2.rst
+++ b/doc/source/whatsnew/v1.0.2.rst
@@ -83,6 +83,10 @@ Bug fixes
 - Fixed bug where :meth:`GroupBy.first` and :meth:`GroupBy.last` would raise a ``TypeError`` when groups contained ``pd.NA`` in a column of object dtype (:issue:`32123`)
 - Fix bug in :meth:`Series.convert_dtypes` for series with mix of integers and strings (:issue:`32117`)
 
+**Rolling**
+
+- Fixed rolling operations with variable window (defined by time duration) on decreasing time index (:issue:`32385`).
+
 .. ---------------------------------------------------------------------------
 
 .. _whatsnew_102.contributors:
diff --git a/pandas/_libs/window/aggregations.pyx b/pandas/_libs/window/aggregations.pyx
index 80b9144042041..a90d2f77e44d1 100644
--- a/pandas/_libs/window/aggregations.pyx
+++ b/pandas/_libs/window/aggregations.pyx
@@ -1013,7 +1013,7 @@ def roll_max_variable(ndarray[float64_t] values, ndarray[int64_t] start,
 def roll_min_fixed(ndarray[float64_t] values, ndarray[int64_t] start,
                    ndarray[int64_t] end, int64_t minp, int64_t win):
     """
-    Moving max of 1d array of any numeric type along axis=0 ignoring NaNs.
+    Moving min of 1d array of any numeric type along axis=0 ignoring NaNs.
 
     Parameters
     ----------
@@ -1030,7 +1030,7 @@ def roll_min_fixed(ndarray[float64_t] values, ndarray[int64_t] start,
 def roll_min_variable(ndarray[float64_t] values, ndarray[int64_t] start,
                       ndarray[int64_t] end, int64_t minp):
     """
-    Moving max of 1d array of any numeric type along axis=0 ignoring NaNs.
+    Moving min of 1d array of any numeric type along axis=0 ignoring NaNs.
 
     Parameters
     ----------
diff --git a/pandas/_libs/window/indexers.pyx b/pandas/_libs/window/indexers.pyx
index 2d01d1964c043..8a1e7feb57ace 100644
--- a/pandas/_libs/window/indexers.pyx
+++ b/pandas/_libs/window/indexers.pyx
@@ -44,6 +44,7 @@ def calculate_variable_window_bounds(
     cdef:
         bint left_closed = False
         bint right_closed = False
+        int index_growth_sign = 1
         ndarray[int64_t, ndim=1] start, end
         int64_t start_bound, end_bound
         Py_ssize_t i, j
@@ -58,6 +59,9 @@ def calculate_variable_window_bounds(
     if closed in ['left', 'both']:
         left_closed = True
 
+    if index[num_values - 1] < index[0]:
+        index_growth_sign = -1
+
     start = np.empty(num_values, dtype='int64')
     start.fill(-1)
     end = np.empty(num_values, dtype='int64')
@@ -78,7 +82,7 @@ def calculate_variable_window_bounds(
         # end is end of slice interval (not including)
         for i in range(1, num_values):
             end_bound = index[i]
-            start_bound = index[i] - window_size
+            start_bound = index[i] - index_growth_sign * window_size
 
             # left endpoint is closed
             if left_closed:
@@ -88,13 +92,13 @@ def calculate_variable_window_bounds(
             # within the constraint
             start[i] = i
             for j in range(start[i - 1], i):
-                if index[j] > start_bound:
+                if (index[j] - start_bound) * index_growth_sign > 0:
                     start[i] = j
                     break
 
             # end bound is previous end
             # or current index
-            if index[end[i - 1]] <= end_bound:
+            if (index[end[i - 1]] - end_bound) * index_growth_sign <= 0:
                 end[i] = i + 1
             else:
                 end[i] = end[i - 1]
diff --git a/pandas/tests/window/test_timeseries_window.py b/pandas/tests/window/test_timeseries_window.py
index 5f5e10b5dd497..0c5289cd78fed 100644
--- a/pandas/tests/window/test_timeseries_window.py
+++ b/pandas/tests/window/test_timeseries_window.py
@@ -709,20 +709,25 @@ def test_rolling_cov_offset(self):
         tm.assert_series_equal(result, expected2)
 
     def test_rolling_on_decreasing_index(self):
-        # GH-19248
+        # GH-19248, GH-32385
         index = [
-            Timestamp("20190101 09:00:00"),
-            Timestamp("20190101 09:00:02"),
-            Timestamp("20190101 09:00:03"),
-            Timestamp("20190101 09:00:05"),
-            Timestamp("20190101 09:00:06"),
+            Timestamp("20190101 09:00:30"),
+            Timestamp("20190101 09:00:27"),
+            Timestamp("20190101 09:00:20"),
+            Timestamp("20190101 09:00:18"),
+            Timestamp("20190101 09:00:10"),
         ]
 
-        df = DataFrame({"column": [3, 4, 4, 2, 1]}, index=reversed(index))
-        result = df.rolling("2s").min()
-        expected = DataFrame(
-            {"column": [3.0, 3.0, 3.0, 2.0, 1.0]}, index=reversed(index)
-        )
+        df = DataFrame({"column": [3, 4, 4, 5, 6]}, index=index)
+        result = df.rolling("5s").min()
+        expected = DataFrame({"column": [3.0, 3.0, 4.0, 4.0, 6.0]}, index=index)
+        tm.assert_frame_equal(result, expected)
+
+    def test_rolling_on_empty(self):
+        # GH-32385
+        df = DataFrame({"column": []}, index=[])
+        result = df.rolling("5s").min()
+        expected = DataFrame({"column": []}, index=[])
         tm.assert_frame_equal(result, expected)
 
     def test_rolling_on_multi_index_level(self):