Two techniques are presented to improve the performance of a binary joint transform correlator system for fingerprint recognition. Use of a Fourier plane binarization threshold that alternates between even and odd rows is shown to reduce noise in the output plane and to improve discrimination between matched and unmatched prints. Periodically correlating with a known standard input and normalizing the output peak intensity prevents correlator performance drift from degrading fingerprint recognition performance. Experimental system performance is assessed statistically in terms of the false pass rate seen when operating with a constant false alarm rate. By applying both techniques, a false pass improvement by a factor of almost 100 is experimentally demonstrated at a false alarm rate of 0.1%.