diff --git a/lir/persistence.py b/lir/persistence.py
index a395b21..1841fd8 100644
--- a/lir/persistence.py
+++ b/lir/persistence.py
@@ -1,4 +1,5 @@
 import pickle
+from os import PathLike
 from pathlib import Path
 
 from lir.aggregation import Aggregation, AggregationData
@@ -26,12 +27,30 @@ def save_model(path: Path, model: LRSystem) -> None:
 class SaveModel(Aggregation):
     """Write the model to a file."""
 
-    def __init__(self, path: Path):
-        self.path = path
+    def __init__(self, output_dir: Path, filename: PathLike | str = 'model.pkl') -> None:
+        """
+        Initialize the aggregation object.
+
+        The model is saved as a pickle file, in a file named `filename`, that is written to a subdirectory of
+        `output_dir`, that is created for each run.
+
+        If `filename` is an absolute path, or if `filename` is relative to `output_dir`, then the model is saved to this
+        file as-is, instead of to a file in a newly created subdirectory.
+
+        :param output_dir: the directory where the model should be written
+        :param filename: the filename to be created for the model
+        """
+        self.output_dir = output_dir
+        self.filename = Path(filename)
 
     def report(self, data: AggregationData) -> None:
-        """Write the trained LR system model to file."""
-        save_model(self.path, data.lrsystem)
+        """Create a directory for the run and write the trained LR system model to file."""
+        if self.filename.is_absolute() or self.filename.is_relative_to(self.output_dir):
+            save_model(self.filename, data.lrsystem)
+        else:
+            dirname = self.output_dir / data.run_name if data.run_name else self.output_dir
+            dirname.mkdir(parents=True, exist_ok=True)
+            save_model(dirname / self.filename, data.lrsystem)
 
 
 @config_parser
@@ -44,4 +63,4 @@ def parse_save_model(config: ContextAwareDict, output_dir: Path) -> SaveModel:
     """
     filename = pop_field(config, 'filename', default='model.pkl', validate=str)
     check_is_empty(config)
-    return SaveModel(output_dir / filename)
+    return SaveModel(output_dir, filename)