Abstract: Techniques for snapshot classification involve generating a plurality of snapshots of a storage system at multiple times; clustering the plurality of snapshots into a first group of snapshots and a second group of snapshots based on data features of the plurality of snapshots; and determining a clean snapshot among the plurality of snapshots based on a comparison between the generation time of the first group of snapshots and the generation time of the second group of snapshots. Accordingly, a clean snapshot and a damaged snapshot can be automatically and quickly distinguished, thereby improving the efficiency of finding a clean snapshot and helping a user recover data quickly.