In two studies, the alternate-form reliability of the Snodgrass picture fragment completion test of implicit memory (Snodgrass, Smith, Feenan, & Corwin, 1987) was examined. In this test, identification thresholds are established for fragmented pictures. The same fragmented pictures are then shown again, intermixed with new fragmented pictures. Implicit memory is indicated by a decrease in identification threshold from the first to second presentation. Alternate-form reliability was low to moderate, depending on the measure used, regardless of the length of the test. A third study showed that explicit memory instructions did not increase the reliability. Recommendations for use of the test in correlational and experimental research are presented.