diff --git a/src/secmlt/adv/evasion/base_evasion_attack.py b/src/secmlt/adv/evasion/base_evasion_attack.py
index 4e291503..1fe8a380 100644
--- a/src/secmlt/adv/evasion/base_evasion_attack.py
+++ b/src/secmlt/adv/evasion/base_evasion_attack.py
@@ -2,7 +2,7 @@
 
 import importlib.util
 from abc import abstractmethod
-from typing import Literal
+from typing import Literal, Union
 
 import torch
 from secmlt.adv.backends import Backends
@@ -144,14 +144,14 @@ def get_backends() -> set[str]:
 class BaseEvasionAttack:
     """Base class for evasion attacks."""
 
-    def __call__(self, model: BaseModel, data_loader: DataLoader) -> DataLoader:
+    def __call__(self, model: Union[BaseModel, list[BaseModel]], data_loader: DataLoader) -> DataLoader:
         """
         Compute the attack against the model, using the input data.
 
         Parameters
         ----------
-        model : BaseModel
-            Model to test.
+        model : BaseModel | list[BaseModel]
+            Model or list of models to test.
         data_loader : DataLoader
             Test dataloader.
 
diff --git a/src/secmlt/adv/evasion/modular_attack.py b/src/secmlt/adv/evasion/modular_attack.py
index 93f2e0f5..963cbf07 100644
--- a/src/secmlt/adv/evasion/modular_attack.py
+++ b/src/secmlt/adv/evasion/modular_attack.py
@@ -158,8 +158,8 @@ def forward_loss(
 
         Parameters
         ----------
-        model : BaseModel
-            Model used by the attack run.
+        model : BaseModel | list[BaseModel]
+            Model(s) used by the attack run.
         x : torch.Tensor
             Input sample.
         target : torch.Tensor
@@ -170,9 +170,17 @@ def forward_loss(
         tuple[torch.Tensor, torch.Tensor]
             Output scores and loss.
         """
-        scores = model.decision_function(x)
-        target = target.to(scores.device)
-        losses = self.loss_function(scores, target)
+        if isinstance(model, list):
+            scores_list = [m.decision_function(x) for m in model]
+            scores = torch.stack(scores_list).mean(dim=0)
+            losses = torch.stack(
+                [self.loss_function(scores, target.to(scores.device)) 
+                    for scores in scores_list]
+            ).mean(dim=0) 
+        else:
+            scores = model.decision_function(x)
+            losses = self.loss_function(scores, 
+                target.to(scores.device))
         return scores, losses
 
     def _run(