We describe an efficient method for assembling short reads into long sequences. In this method, a hashing technique is used to compute overlaps between short reads, allowing base mismatches in the overlaps. Then an overlap graph is constructed, with each vertex representing a read and each edge representing an overlap. The overlap graph is explored by graph algorithms to find unique paths of reads representing contigs. The consensus sequence of each contig is constructed by computing alignments of multiple reads without gaps. This strategy has been implemented as a short read assembly program called PCAP.Solexa. We also describe how to use PCAP. Solexa in assembly of short reads.