diff --git a/pytm/json.py b/pytm/json.py
index e9a1deb..d202b71 100644
--- a/pytm/json.py
+++ b/pytm/json.py
@@ -1,13 +1,15 @@
 import json
 
 from .tm import TM
+from .base import DataSet
 from .boundary import Boundary
+from .data import Data
 from .dataflow import Dataflow
 from .asset import Asset, Server, ExternalEntity, Lambda, LLM
 from .datastore import Datastore
 from .actor import Actor
 from .process import Process, SetOfProcesses
-from .enums import Action
+from .enums import Action, Classification, Lifetime
 
 _ELEMENT_CLASSES = {
     "Asset": Asset,
@@ -24,36 +26,30 @@
 
 def loads(s):
     """Load a TM object from a JSON string *s*."""
-    result = json.loads(s, object_hook=decode)
-    if not isinstance(result, TM):
-        raise ValueError("Failed to decode JSON input as TM")
-    return result
+    result = json.loads(s)
+    return _decode(result)
 
 
 def load(fp):
     """Load a TM object from an open file containing JSON."""
-    result = json.load(fp, object_hook=decode)
-    if not isinstance(result, TM):
-        raise ValueError("Failed to decode JSON input as TM")
-    return result
+    result = json.load(fp)
+    return _decode(result)
 
 
-def decode(data):
-    if "elements" not in data and "flows" not in data and "boundaries" not in data:
-        return data
+def _decode(flat):
+    boundaries = _decode_boundaries(flat.pop("boundaries", []))
+    data = _decode_data(flat.pop("data", []))
+    elements = _decode_elements(flat.pop("elements", []), boundaries)
+    _decode_flows(flat.pop("flows", []), elements, data)
 
-    boundaries = decode_boundaries(data.pop("boundaries", []))
-    elements = decode_elements(data.pop("elements", []), boundaries)
-    decode_flows(data.pop("flows", []), elements)
-
-    if "name" not in data:
+    if "name" not in flat:
         raise ValueError("name property missing for threat model")
-    if "onDuplicates" in data:
-        data["onDuplicates"] = Action(data["onDuplicates"])
-    return TM(data.pop("name"), **data)
+    if "onDuplicates" in flat:
+        flat["onDuplicates"] = Action(flat["onDuplicates"])
+    return TM(flat.pop("name"), **flat)
 
 
-def decode_boundaries(flat):
+def _decode_boundaries(flat):
     boundaries = {}
     refs = {}
     for i, e in enumerate(flat):
@@ -74,7 +70,28 @@ def decode_boundaries(flat):
     return boundaries
 
 
-def decode_elements(flat, boundaries):
+def _decode_data(flat):
+    data = {}
+    for i, e in enumerate(flat):
+        name = e.pop("name", None)
+        if name is None:
+            raise ValueError(f"name property missing in data {i}")
+
+        classification_name = e.pop("classification", None)
+        if classification_name:
+            e["classification"] = Classification[classification_name]
+
+        lifetime_name = e.pop("lifetime", None)
+        if lifetime_name:
+            e["lifetime"] = Lifetime[lifetime_name]
+
+        d = Data(name, **e)
+        data[name] = d
+
+    return data
+
+
+def _decode_elements(flat, boundaries):
     elements = {}
     for i, e in enumerate(flat):
         class_name = e.pop("__class__", "Asset")
@@ -96,7 +113,7 @@ def decode_elements(flat, boundaries):
     return elements
 
 
-def decode_flows(flat, elements):
+def _decode_flows(flat, elements, data):
     for i, e in enumerate(flat):
         name = e.pop("name", None)
         if name is None:
@@ -111,4 +128,13 @@ def decode_flows(flat, elements):
         if e["sink"] not in elements:
             raise ValueError(f"dataflow {name} references invalid sink {e['sink']}")
         sink = elements[e.pop("sink")]
+
+        if "data" in e:
+            dataset = DataSet()
+            for data_name in e["data"]:
+                if data_name not in data:
+                    raise ValueError(f"dataflow {name} references invalid data {data_name}")
+                dataset.add(data[data_name])
+            e["data"] = dataset
+
         Dataflow(source, sink, name, **e)
diff --git a/tests/input.json b/tests/input.json
index 0cdcff8..798a284 100644
--- a/tests/input.json
+++ b/tests/input.json
@@ -11,6 +11,13 @@
       "name": "Server/DB"
     }
   ],
+  "data": [
+    {
+      "name": "Password",
+      "classification": "SECRET",
+      "lifetime": "LONG"
+    }
+  ],
   "elements": [
     {
       "__class__": "Actor",
@@ -32,6 +39,9 @@
       "name": "Request",
       "source": "User",
       "sink": "Web Server",
+      "data": [
+        "Password"
+      ],
       "note": "bbb"
     },
     {
diff --git a/tests/test_pytmfunc.py b/tests/test_pytmfunc.py
index c0696be..00dac53 100644
--- a/tests/test_pytmfunc.py
+++ b/tests/test_pytmfunc.py
@@ -426,6 +426,14 @@ def test_json_loads(self):
         ]
         assert [f.name for f in tm._flows] == ["Request", "Insert", "Select", "Response"]
 
+        assert [d.model_dump(include=["name", "classification", "lifetime"]) for d in tm._data] == [
+            {
+                "name": "Password",
+                "classification": Classification.SECRET,
+                "lifetime": Lifetime.LONG,
+            },
+        ]
+
     @pytest.mark.parametrize(
         "class_name,expected_type",
         [