diff --git a/swh/provenance/archive.py b/swh/provenance/archive.py
--- a/swh/provenance/archive.py
+++ b/swh/provenance/archive.py
@@ -15,7 +15,7 @@
 class ArchiveInterface(Protocol):
     storage: StorageInterface
 
-    def directory_ls(self, id: Sha1Git) -> Iterable[Dict[str, Any]]:
+    def directory_ls(self, id: Sha1Git, minsize: int = 0) -> Iterable[Dict[str, Any]]:
         """List entries for one directory.
 
         Args:
diff --git a/swh/provenance/cli.py b/swh/provenance/cli.py
--- a/swh/provenance/cli.py
+++ b/swh/provenance/cli.py
@@ -130,6 +130,7 @@
 @click.option("-l", "--limit", type=int)
 @click.option("-m", "--min-depth", default=1, type=int)
 @click.option("-r", "--reuse", default=True, type=bool)
+@click.option("-s", "--min-size", default=0, type=int)
 @click.pass_context
 def iter_revisions(
     ctx: click.core.Context,
@@ -138,8 +139,8 @@
     limit: Optional[int],
     min_depth: int,
     reuse: bool,
+    min_size: int,
 ) -> None:
-    # TODO: add file size filtering
     """Process a provided list of revisions."""
     from . import get_archive, get_provenance
     from .revision import CSVRevisionIterator, revision_add
@@ -157,6 +158,7 @@
                 trackall=track_all,
                 lower=reuse,
                 mindepth=min_depth,
+                minsize=min_size,
             )
 
 
diff --git a/swh/provenance/graph.py b/swh/provenance/graph.py
--- a/swh/provenance/graph.py
+++ b/swh/provenance/graph.py
@@ -176,6 +176,7 @@
     provenance: ProvenanceInterface,
     revision: RevisionEntry,
     directory: DirectoryEntry,
+    minsize: int = 0,
 ) -> IsochroneNode:
     assert revision.date is not None
     assert revision.root == directory.id
@@ -208,7 +209,7 @@
             # Pre-query all known dates for directories in the current directory
             # for the provenance object to have them cached and (potentially) improve
             # performance.
-            current.entry.retrieve_children(archive)
+            current.entry.retrieve_children(archive, minsize=minsize)
             ddates = provenance.directory_get_dates_in_isochrone_frontier(
                 current.entry.dirs
             )
diff --git a/swh/provenance/model.py b/swh/provenance/model.py
--- a/swh/provenance/model.py
+++ b/swh/provenance/model.py
@@ -86,11 +86,11 @@
         self._files: Optional[List[FileEntry]] = None
         self._dirs: Optional[List[DirectoryEntry]] = None
 
-    def retrieve_children(self, archive: ArchiveInterface) -> None:
+    def retrieve_children(self, archive: ArchiveInterface, minsize: int = 0) -> None:
         if self._files is None and self._dirs is None:
             self._files = []
             self._dirs = []
-            for child in archive.directory_ls(self.id):
+            for child in archive.directory_ls(self.id, minsize=minsize):
                 if child["type"] == "dir":
                     self._dirs.append(
                         DirectoryEntry(child["target"], name=child["name"])
diff --git a/swh/provenance/postgresql/archive.py b/swh/provenance/postgresql/archive.py
--- a/swh/provenance/postgresql/archive.py
+++ b/swh/provenance/postgresql/archive.py
@@ -22,14 +22,13 @@
         )
         self.conn = conn
 
-    def directory_ls(self, id: Sha1Git) -> Iterable[Dict[str, Any]]:
-        entries = self._directory_ls(id)
+    def directory_ls(self, id: Sha1Git, minsize: int = 0) -> Iterable[Dict[str, Any]]:
+        entries = self._directory_ls(id, minsize=minsize)
         yield from entries
 
     @lru_cache(maxsize=100000)
     @statsd.timed(metric=ARCHIVE_DURATION_METRIC, tags={"method": "directory_ls"})
-    def _directory_ls(self, id: Sha1Git) -> List[Dict[str, Any]]:
-        # TODO: add file size filtering
+    def _directory_ls(self, id: Sha1Git, minsize: int = 0) -> List[Dict[str, Any]]:
         with self.conn.cursor() as cursor:
             cursor.execute(
                 """
@@ -49,7 +48,9 @@
                             c.sha1_git
                         FROM ls_f
                         LEFT JOIN directory_entry_file e ON ls_f.entry_id=e.id
-                        INNER JOIN content c ON e.target=c.sha1_git)
+                        INNER JOIN content c ON e.target=c.sha1_git
+                        WHERE c.length >= %s
+                    )
                     SELECT * FROM known_contents
                     UNION
                     (SELECT 'file'::directory_entry_type AS type, e.target, e.name,
@@ -61,10 +62,11 @@
                             SELECT 1 FROM known_contents
                                 WHERE known_contents.sha1_git=e.target
                         )
+                        AND c.length >= %s
                     )
                 )
                 """,
-                (id,),
+                (id, minsize, minsize),
             )
             return [
                 {"type": row[0], "target": row[1], "name": row[2]} for row in cursor
@@ -84,7 +86,6 @@
                 """,
                 (id,),
             )
-            # There should be at most one row anyway
             yield from (row[0] for row in cursor)
 
     @statsd.timed(metric=ARCHIVE_DURATION_METRIC, tags={"method": "snapshot_get_heads"})
diff --git a/swh/provenance/revision.py b/swh/provenance/revision.py
--- a/swh/provenance/revision.py
+++ b/swh/provenance/revision.py
@@ -59,6 +59,7 @@
     trackall: bool = True,
     lower: bool = True,
     mindepth: int = 1,
+    minsize: int = 0,
     commit: bool = True,
 ) -> None:
     for revision in revisions:
@@ -72,8 +73,8 @@
                 provenance,
                 revision,
                 DirectoryEntry(revision.root),
+                minsize=minsize,
             )
-            # TODO: add file size filtering
             revision_process_content(
                 archive,
                 provenance,
@@ -82,6 +83,7 @@
                 trackall=trackall,
                 lower=lower,
                 mindepth=mindepth,
+                minsize=minsize,
             )
     if commit:
         provenance.flush()
@@ -96,6 +98,7 @@
     trackall: bool = True,
     lower: bool = True,
     mindepth: int = 1,
+    minsize: int = 0,
 ) -> None:
     assert revision.date is not None
     provenance.revision_add(revision)
@@ -132,7 +135,9 @@
                     provenance.directory_add_to_revision(
                         revision, current.entry, current.path
                     )
-                    flatten_directory(archive, provenance, current.entry)
+                    flatten_directory(
+                        archive, provenance, current.entry, minsize=minsize
+                    )
             else:
                 # If current node is an invalidated frontier, update its date for future
                 # revisions to get the proper value.
@@ -158,6 +163,7 @@
     archive: ArchiveInterface,
     provenance: ProvenanceInterface,
     directory: DirectoryEntry,
+    minsize: int = 0,
 ) -> None:
     """Recursively retrieve all the files of 'directory' and insert them in the
     'provenance' database in the 'content_to_directory' table.
@@ -165,7 +171,7 @@
     stack = [(directory, b"")]
     while stack:
         current, prefix = stack.pop()
-        current.retrieve_children(archive)
+        current.retrieve_children(archive, minsize=minsize)
         for f_child in current.files:
             # Add content to the directory with the computed prefix.
             provenance.content_add_to_directory(directory, f_child, prefix)
diff --git a/swh/provenance/storage/archive.py b/swh/provenance/storage/archive.py
--- a/swh/provenance/storage/archive.py
+++ b/swh/provenance/storage/archive.py
@@ -18,14 +18,16 @@
         self.storage = storage
 
     @statsd.timed(metric=ARCHIVE_DURATION_METRIC, tags={"method": "directory_ls"})
-    def directory_ls(self, id: Sha1Git) -> Iterable[Dict[str, Any]]:
-        # TODO: add file size filtering
+    def directory_ls(self, id: Sha1Git, minsize: int = 0) -> Iterable[Dict[str, Any]]:
         for entry in self.storage.directory_ls(id):
-            yield {
-                "name": entry["name"],
-                "target": entry["target"],
-                "type": entry["type"],
-            }
+            if entry["type"] == "dir" or (
+                entry["type"] == "file" and entry["length"] >= minsize
+            ):
+                yield {
+                    "name": entry["name"],
+                    "target": entry["target"],
+                    "type": entry["type"],
+                }
 
     @statsd.timed(
         metric=ARCHIVE_DURATION_METRIC, tags={"method": "revision_get_parents"}