diff --git a/mapchete/_core.py b/mapchete/_core.py
index 620957e6..bde9f21c 100644
--- a/mapchete/_core.py
+++ b/mapchete/_core.py
@@ -1,13 +1,13 @@
 """Main module managing processes."""
 
 from cachetools import LRUCache
-import concurrent.futures
 import logging
 import multiprocessing
 import threading
 
 from mapchete.config import MapcheteConfig
 from mapchete.errors import MapcheteNodataTile
+from mapchete.io import process_tiles_exist
 from mapchete._processing import _run_on_single_tile, _run_area, ProcessInfo, TileProcess
 from mapchete.tile import count_tiles
 from mapchete._timer import Timer
@@ -145,16 +145,10 @@ def skip_tiles(self, tiles=None):
         ------
         tuples : (tile, skip)
         """
-        def _skip(config, tile):
-            return tile, config.output_reader.tiles_exist(tile)
-
         # only check for existing output in "continue" mode
         if self.config.mode == "continue":
-            with concurrent.futures.ThreadPoolExecutor() as executor:
-                for future in concurrent.futures.as_completed(
-                    (executor.submit(_skip, self.config, tile) for tile in tiles)
-                ):
-                    yield future.result()
+            yield from process_tiles_exist(config=self.config, process_tiles=tiles)
+        # otherwise don't skip tiles
         else:
             for tile in tiles:
                 yield (tile, False)
diff --git a/mapchete/io/__init__.py b/mapchete/io/__init__.py
index e6372b2f..5bbda3eb 100644
--- a/mapchete/io/__init__.py
+++ b/mapchete/io/__init__.py
@@ -15,6 +15,7 @@
 from mapchete.io._path import (
     path_is_remote,
     path_exists,
+    process_tiles_exist,
     absolute_path,
     relative_path,
     makedirs,
@@ -28,6 +29,7 @@
     "tile_to_zoom_level",
     "path_is_remote",
     "path_exists",
+    "process_tiles_exist",
     "absolute_path",
     "relative_path",
     "makedirs",
diff --git a/mapchete/io/_path.py b/mapchete/io/_path.py
index e9a7b7a1..de719502 100644
--- a/mapchete/io/_path.py
+++ b/mapchete/io/_path.py
@@ -1,9 +1,14 @@
+from collections import defaultdict
+import concurrent.futures
+import logging
 import os
 from urllib.request import urlopen
 from urllib.error import HTTPError
 
 from mapchete.io._misc import get_boto3_bucket
 
+logger = logging.getLogger(__name__)
+
 
 def path_is_remote(path, s3=True):
     """
@@ -112,3 +117,94 @@ def makedirs(path):
             os.makedirs(path)
         except OSError:
             pass
+
+
+def process_tiles_exist(config=None, process_tiles=None):
+    """
+    Yield process tiles and whether their output already exists or not.
+
+    The S3 part of the function are loosely inspired by
+    https://alexwlchan.net/2019/07/listing-s3-keys/
+
+    Parameters
+    ----------
+    config : mapchete.config.MapcheteConfig
+    process_tiles : iterator
+
+    Yields
+    ------
+    tuple : (process_tile, exists)
+    """
+    basepath = config.output_reader.path
+
+    # only on TileDirectories on S3
+    if (
+        not config.output_reader.path.endswith(config.output_reader.file_extension) and
+        basepath.startswith("s3://")
+    ):
+        basekey = "/".join(basepath.split("/")[3:])
+        bucket = basepath.split("/")[2]
+        import boto3
+        s3 = boto3.client("s3")
+        paginator = s3.get_paginator("list_objects_v2")
+
+        # make process tiles unique
+        process_tiles = set(process_tiles)
+        # determine current zoom
+        zoom = next(iter(process_tiles)).zoom
+        # get all output tiles for process tiles
+        output_tiles = set(
+            [
+                t
+                for process_tile in process_tiles
+                for t in config.output_pyramid.intersecting(process_tile)
+            ]
+        )
+        # create a mapping between paths and process_tiles
+        paths = dict()
+        # group process_tiles by row
+        rowgroups = defaultdict(list)
+        # remember already yielded process_tiles
+        yielded = set()
+        for output_tile in output_tiles:
+            if output_tile.zoom != zoom:
+                raise ValueError("tiles of different zoom levels cannot be mixed")
+            path = config.output_reader.get_path(output_tile)
+            process_tile = config.process_pyramid.intersecting(output_tile)[0]
+            paths[path] = process_tile
+            rowgroups[process_tile.row].append(process_tile)
+        # use prefix until row, page through api results
+        for row, tiles in rowgroups.items():
+            logger.debug("check existing tiles in row %s" % row)
+            prefix = os.path.join(*[basekey, str(zoom), str(row)])
+            logger.debug(
+                "read keys %s*" % os.path.join("s3://" + bucket, prefix)
+            )
+
+            for page in paginator.paginate(Bucket=bucket, Prefix=prefix):
+                logger.debug("read next page")
+                try:
+                    contents = page["Contents"]
+                except KeyError:
+                    break
+                for obj in contents:
+                    path = obj["Key"]
+                    # get matching process_tile
+                    process_tile = paths[os.path.join("s3://" + bucket, path)]
+                    # store and yield process tile if it was not already yielded
+                    if process_tile not in yielded:
+                        yielded.add(process_tile)
+                        yield (process_tile, True)
+
+        # finally, yield all process tiles which were not yet yielded as False
+        for process_tile in process_tiles.difference(yielded):
+            yield (process_tile, False)
+
+    else:
+        def _exists(tile):
+            return (tile, config.output_reader.tiles_exist(tile))
+        with concurrent.futures.ThreadPoolExecutor() as executor:
+            for future in concurrent.futures.as_completed(
+                (executor.submit(_exists, tile) for tile in process_tiles)
+            ):
+                yield future.result()