diff --git a/osmnx/convert.py b/osmnx/convert.py
index 27cc65ab..2909bc35 100644
--- a/osmnx/convert.py
+++ b/osmnx/convert.py
@@ -222,26 +222,14 @@ def _make_edge_geometry(
     raise ValueError(msg)
 
 
-def graph_from_gdfs(
+def _validate_node_edge_gdfs(
     gdf_nodes: gpd.GeoDataFrame,
     gdf_edges: gpd.GeoDataFrame,
-    *,
-    graph_attrs: dict[str, Any] | None = None,
-) -> nx.MultiDiGraph:
+) -> None:
     """
-    Convert node and edge GeoDataFrames to a MultiDiGraph.
-
-    This function is the inverse of `graph_to_gdfs` and is designed to work in
-    conjunction with it. However, you can convert arbitrary node and edge
-    GeoDataFrames as long as 1) `gdf_nodes` is uniquely indexed by `osmid`, 2)
-    `gdf_nodes` contains `x` and `y` coordinate columns representing node
-    geometries, 3) `gdf_edges` is uniquely multi-indexed by `(u, v, key)`
-    (following normal MultiDiGraph structure). This allows you to load any
-    node/edge Shapefiles or GeoPackage layers as GeoDataFrames then convert
-    them to a MultiDiGraph for network analysis.
+    Validate that node/edge GeoDataFrames can be converted to a MultiDiGraph.
 
-    Note that any `geometry` attribute on `gdf_nodes` is discarded, since `x`
-    and `y` provide the necessary node geometry information instead.
+    Raises a `ValueError` if validation fails.
 
     Parameters
     ----------
@@ -250,13 +238,10 @@ def graph_from_gdfs(
     gdf_edges
         GeoDataFrame of graph edges uniquely multi-indexed by `(u, v, key)`.
     graph_attrs
-        The new `G.graph` attribute dictionary. If None, use `gdf_edges`'s CRS
-        as the only graph-level attribute (`gdf_edges` must have its `crs`
-        attribute set).
 
     Returns
     -------
-    G
+    None
     """
     # ensure gdf_nodes contains x and y columns representing node geometries
     if not ("x" in gdf_nodes.columns and "y" in gdf_nodes.columns):  # pragma: no cover
@@ -278,12 +263,9 @@ def graph_from_gdfs(
         msg = "`gdf_edges` must be multi-indexed by `(u, v, key)`."
         raise ValueError(msg)
 
-    # drop geometry column from gdf_nodes (since we use x and y for geometry
-    # information), but warn the user if the geometry values differ from the
-    # coordinates in the x and y columns. this results in a df instead of gdf.
-    if gdf_nodes.active_geometry_name is None:  # pragma: no cover
-        df_nodes = pd.DataFrame(gdf_nodes)
-    else:
+    # warn user if geometry values differ from coordinates in x/y columns,
+    # because we discard the geometry column
+    if gdf_nodes.active_geometry_name is not None:  # pragma: no cover
         msg = (
             "Discarding the `gdf_nodes` 'geometry' column, though its values "
             "differ from the coordinates in the 'x' and 'y' columns."
@@ -297,7 +279,53 @@ def graph_from_gdfs(
         except ValueError:  # pragma: no cover
             # warn if geometry column contains non-point geometry types
             warn(msg, category=UserWarning, stacklevel=2)
-        df_nodes = gdf_nodes.drop(columns=gdf_nodes.geometry.name)
+
+
+def graph_from_gdfs(
+    gdf_nodes: gpd.GeoDataFrame,
+    gdf_edges: gpd.GeoDataFrame,
+    *,
+    graph_attrs: dict[str, Any] | None = None,
+) -> nx.MultiDiGraph:
+    """
+    Convert node and edge GeoDataFrames to a MultiDiGraph.
+
+    This function is the inverse of `graph_to_gdfs` and is designed to work in
+    conjunction with it. However, you can convert arbitrary node and edge
+    GeoDataFrames as long as 1) `gdf_nodes` is uniquely indexed by `osmid`, 2)
+    `gdf_nodes` contains `x` and `y` coordinate columns representing node
+    geometries, 3) `gdf_edges` is uniquely multi-indexed by `(u, v, key)`
+    (following normal MultiDiGraph structure). This allows you to load any
+    node/edge Shapefiles or GeoPackage layers as GeoDataFrames then convert
+    them to a MultiDiGraph for network analysis.
+
+    Note that any `geometry` attribute on `gdf_nodes` is discarded, since `x`
+    and `y` provide the necessary node geometry information instead.
+
+    Parameters
+    ----------
+    gdf_nodes
+        GeoDataFrame of graph nodes uniquely indexed by `osmid`.
+    gdf_edges
+        GeoDataFrame of graph edges uniquely multi-indexed by `(u, v, key)`.
+    graph_attrs
+        The new `G.graph` attribute dictionary. If None, use `gdf_edges`'s CRS
+        as the only graph-level attribute (`gdf_edges` must have its `crs`
+        attribute set).
+
+    Returns
+    -------
+    G
+    """
+    _validate_node_edge_gdfs(gdf_nodes, gdf_edges)
+
+    # drop geometry column from gdf_nodes (since we use x and y for geometry
+    # information), but warn the user if the geometry values differ from the
+    # coordinates in the x and y columns. this results in a df instead of gdf.
+    if gdf_nodes.active_geometry_name is None:  # pragma: no cover
+        df_nodes = pd.DataFrame(gdf_nodes)
+    else:
+        df_nodes = gdf_nodes.drop(columns=gdf_nodes.active_geometry_name)
 
     # create graph and add graph-level attribute dict
     if graph_attrs is None: