diff --git a/eqcorrscan/core/match_filter/detection.py b/eqcorrscan/core/match_filter/detection.py
index 9ac074dd..4b11f380 100644
--- a/eqcorrscan/core/match_filter/detection.py
+++ b/eqcorrscan/core/match_filter/detection.py
@@ -397,7 +397,8 @@ def extract_stream(self, stream, length, prepick, all_vert=False,
                 pick = [p for p in pick
                         if p.waveform_id.channel_code == channel]
             if len(pick) == 0:
-                Logger.info("No pick for {0}.{1}".format(station, channel))
+                Logger.info(
+                    "No pick for {0}.{1}".format(station, channel))
                 continue
             elif len(pick) > 1:
                 Logger.info(
@@ -406,13 +407,28 @@ def extract_stream(self, stream, length, prepick, all_vert=False,
                 pick.sort(key=lambda p: p.time)
             pick = pick[0]
             cut_start = pick.time - prepick
-            cut_end = cut_start + length
-            _st = _st.slice(starttime=cut_start, endtime=cut_end).copy()
-            # Minimum length check
+            # Find nearest sample to avoid  to too-short length - see #573
             for tr in _st:
-                if abs((tr.stats.endtime - tr.stats.starttime) -
+                sample_offset = (cut_start -
+                                 tr.stats.starttime) * tr.stats.sampling_rate
+                Logger.debug(
+                    f"Sample offset for slice on {tr.id}: {sample_offset}")
+                sample_offset //= 1
+                # If the sample offset is not a whole number, always take the
+                # sample before that requested
+                _tr_cut_start = tr.stats.starttime + (
+                    sample_offset * tr.stats.delta)
+                _tr_cut_end = _tr_cut_start + length
+                Logger.debug(
+                    f"Trimming {tr.id} between {_tr_cut_end} "
+                    f"and {_tr_cut_end}.")
+                _tr = tr.slice(_tr_cut_start, _tr_cut_end).copy()
+                Logger.debug(
+                    f"Length: {(_tr.stats.endtime - _tr.stats.starttime)}")
+                Logger.debug(f"Requested length: {length}")
+                if abs((_tr.stats.endtime - _tr.stats.starttime) -
                        length) < tr.stats.delta:
-                    cut_stream += tr
+                    cut_stream += _tr
                 else:
                     Logger.info(
                         "Insufficient data length for {0}".format(tr.id))
diff --git a/eqcorrscan/tests/lag_calc_test.py b/eqcorrscan/tests/lag_calc_test.py
index 97f33c7c..4a660a3f 100644
--- a/eqcorrscan/tests/lag_calc_test.py
+++ b/eqcorrscan/tests/lag_calc_test.py
@@ -26,7 +26,6 @@ class SyntheticTests(unittest.TestCase):
     def setUpClass(cls):
         np.random.seed(999)
         print("Setting up class")
-        np.random.seed(999)
         samp_rate = 50
         t_length = .75
         # Make some synthetic templates
@@ -34,6 +33,7 @@ def setUpClass(cls):
             nsta=5, ntemplates=5, nseeds=10, samp_rate=samp_rate,
             t_length=t_length, max_amp=10, max_lag=15, phaseout="both",
             jitter=0, noise=False, same_phase=True)
+        print("Made synthetic data")
         # Rename channels
         channel_mapper = {"SYN_Z": "HHZ", "SYN_H": "HHN"}
         for tr in data:
@@ -44,6 +44,7 @@ def setUpClass(cls):
         party = Party()
         t = 0
         data_start = data[0].stats.starttime
+        print("Making party")
         for template, template_seeds in zip(templates, seeds):
             template_name = "template_{0}".format(t)
             detections = []
@@ -68,6 +69,8 @@ def setUpClass(cls):
             family = Family(template=_template, detections=detections)
             party += family
             t += 1
+            print(f"Made template {template_name}")
+        print("Made party")
         cls.party = party
         cls.data = data
         cls.t_length = t_length
diff --git a/eqcorrscan/tests/matched_filter/match_filter_test.py b/eqcorrscan/tests/matched_filter/match_filter_test.py
index fd9cb3e1..9f8ebca1 100644
--- a/eqcorrscan/tests/matched_filter/match_filter_test.py
+++ b/eqcorrscan/tests/matched_filter/match_filter_test.py
@@ -1606,6 +1606,52 @@ def test_family_catalogs(self):
         family.detections.append(additional_detection)
         self.assertEqual(family.catalog, get_catalog(family.detections))
 
+    def test_detection_extract_stream(self):
+        # Create simple synthetic stream
+        traces = []
+        pick_sids = {p.waveform_id.get_seed_string() for f in self.party
+                     for d in f for p in d.event.picks}
+        pick_times = sorted([p.time for f in self.party for d in f
+                             for p in d.event.picks])
+        first_pick, last_pick = pick_times[0], pick_times[-1]
+        delta = self.party[0].template.st[0].stats.delta
+        n_samples = int(((last_pick - first_pick) + 120) / delta)
+        data = np.arange(n_samples, dtype=int)
+        for sid in pick_sids:
+            tr = Trace(data=data.copy())
+            tr.stats.starttime = first_pick - 60
+            n, s, l, c = sid.split('.')
+            tr.stats.delta = delta
+            tr.stats.network = n
+            tr.stats.station = s
+            tr.stats.location = l
+            tr.stats.channel = c
+            traces.append(tr)
+        st = Stream(traces=traces)
+
+        # Test straightforward extraction
+        detection = self.party[0][0]
+        length, pre_pick = 40.0, 1 / delta
+        for shift in range(6):
+            pre_pick -= shift * (delta / 6)  # Sub-sample shifting
+            expected_starts = {
+                p.waveform_id.get_seed_string(): p.time - pre_pick
+                for p in detection.event.picks}
+            cut_st = detection.extract_stream(
+                stream=st, length=length, prepick=pre_pick)
+            for sid, expected_start in expected_starts.items():
+                Logger.debug(f"Checking for {sid}")
+                tr = cut_st.select(id=sid)
+                # Check that we get a returned trace
+                self.assertTrue(len(tr), 1)
+                tr = tr[0]
+                # Check that start is within one sample of the expected start
+                self.assertLess(abs(tr.stats.starttime - expected_start),
+                                delta)
+                # Check that the length is correct
+                returned_length = tr.stats.endtime - tr.stats.starttime
+                self.assertEqual(length, returned_length)
+
 
 class TestTemplateGrouping(unittest.TestCase):
     @classmethod