The widespread use of microarray technology and sequencing of genomes has made it increasingly possible to study the cellular sub-systems of organisms. Computational techniques applied to sequence data annotated with ontologies such as the Gene Ontology (GO) aid in understanding regulatory networks of genes. An important related problem is the estimation of the similarity between gene products based on their annotations. We present an approach to compute gene product similarity that takes into account both the hierarchical nature of GO and the co-occurrence of GO terms in annotations. It also accounts for differences in the cardinality of annotations and differences in the frequency of usage of GO terms. We demonstrate the validity of the metric by computing the similarity between gene products in several different contexts. These include the analysis of similarity within a specific signaling pathway, between proteins constituting a sequence family and the comparative evaluation of different clusterings of microarray data.