diff --git a/robosuite/wrappers/domain_randomization_wrapper.py b/robosuite/wrappers/domain_randomization_wrapper.py
index ce27582663..27365fd638 100644
--- a/robosuite/wrappers/domain_randomization_wrapper.py
+++ b/robosuite/wrappers/domain_randomization_wrapper.py
@@ -207,16 +207,17 @@ def reset(self):
         # normal env reset
         ret = super().reset()
 
+        # a hard reset frees the old sim and builds a new one, so point the
+        # modders at it before reading any model parameters from them
+        for modder in self.modders:
+            modder.update_sim(self.env.sim)
+
         # save the original env parameters
         self.save_default_domain()
 
         # reset counter for doing domain randomization at a particular frequency
         self.step_counter = 0
 
-        # update sims
-        for modder in self.modders:
-            modder.update_sim(self.env.sim)
-
         if self.randomize_on_reset:
             # domain randomize + regenerate observation
             self.randomize_domain()
diff --git a/tests/test_environments/test_domain_randomization.py b/tests/test_environments/test_domain_randomization.py
new file mode 100644
index 0000000000..e64d283356
--- /dev/null
+++ b/tests/test_environments/test_domain_randomization.py
@@ -0,0 +1,41 @@
+"""
+Test that DomainRandomizationWrapper survives a hard reset.
+
+When an environment uses the "mujoco" renderer, a hard reset frees the old MjSim
+and builds a new one. The wrapper's modders keep a reference to the freed sim, so
+saving the default domain must rebind them to the new sim first. Otherwise the
+second reset raises `AttributeError: 'MjSim' object has no attribute 'model'`.
+
+See https://github.com/ARISE-Initiative/robosuite/issues/426.
+"""
+import numpy as np
+
+import robosuite as suite
+from robosuite.wrappers import DomainRandomizationWrapper
+
+
+def test_domain_randomization_hard_reset():
+
+    env = suite.make(
+        env_name="Lift",
+        robots="Panda",
+        has_renderer=False,
+        has_offscreen_renderer=False,
+        use_camera_obs=False,
+        renderer="mujoco",
+    )
+    # randomize_color needs mujoco==3.1.1, so leave it off here
+    env = DomainRandomizationWrapper(env, randomize_color=False)
+
+    # the first reset builds the initial sim; each later hard reset frees it and
+    # builds a new one, which is what used to break the modders
+    for _ in range(3):
+        env.reset()
+        env.step(np.zeros(env.action_dim))
+
+    env.close()
+
+
+if __name__ == "__main__":
+
+    test_domain_randomization_hard_reset()