diff --git a/klamath/library.py b/klamath/library.py
index bbe8bee..1318854 100644
--- a/klamath/library.py
+++ b/klamath/library.py
@@ -1,16 +1,17 @@
 """
 File-level read/write functionality.
 """
-from typing import List, Dict, Tuple, Optional, BinaryIO, TypeVar, Type
+from typing import List, Dict, Tuple, Optional, BinaryIO, TypeVar, Type, MutableMapping
 import io
 from datetime import datetime
 from dataclasses import dataclass
+from collections import defaultdict
 
 from .basic import KlamathError
 from .record import Record
 
 from .records import HEADER, BGNLIB, ENDLIB, UNITS, LIBNAME
-from .records import BGNSTR, STRNAME, ENDSTR
+from .records import BGNSTR, STRNAME, ENDSTR, SNAME, COLROW, ENDEL
 from .records import BOX, BOUNDARY, NODE, PATH, TEXT, SREF, AREF
 from .elements import Element, Reference, Text, Box, Boundary, Path, Node
 
@@ -183,3 +184,50 @@ def read_elements(stream: BinaryIO) -> List[Element]:
             stream.seek(size, io.SEEK_CUR)
         size, tag = Record.read_header(stream)
     return data
+
+
+def scan_hierarchy(stream: BinaryIO) -> Dict[bytes, Dict[bytes, int]]:
+    """
+    Scan through a GDS file, building a table of instance counts
+      `{b'structure_name': {b'ref_name': count}}`.
+
+    This is intended to provide a fast overview of the file's
+     contents without performing a full read of all elements.
+
+    Args:
+        stream: Seekable stream to read from. Should be positioned
+                before the first structure record, but possibly
+                already past the file header.
+    """
+    structures = {}
+
+    ref_name = None
+    ref_count = None
+    cur_structure: MutableMapping[bytes, int] = defaultdict(int)
+    size, tag = Record.read_header(stream)
+    while tag != ENDLIB.tag:
+        if tag == BGNSTR.tag:
+            stream.seek(size, io.SEEK_CUR)
+            name = STRNAME.read(stream)
+            if name in structures:
+                raise KlamathError(f'Duplicate structure name: {name!r}')
+            cur_structure = defaultdict(int)
+            structures[name] = cur_structure
+            ref_name = None
+            ref_count = None
+        elif tag == SNAME.tag:
+            ref_name = SNAME.read_data(stream, size)
+        elif tag == COLROW.tag:
+            colrow = COLROW.read_data(stream, size)
+            ref_count = colrow[0] * colrow[1]
+        elif tag == ENDEL.tag:
+            if ref_count is None:
+                ref_count = 1
+            assert(ref_name is not None)
+            cur_structure[ref_name] += ref_count
+        else:
+            stream.seek(size, io.SEEK_CUR)
+        size, tag = Record.read_header(stream)
+
+    dict_structures = {key: dict(value) for key, value in structures.items()}
+    return dict_structures