NCBI's Homologene is one source, listing pre-calculated orthologs.

Another method is to employ the "mutual best hit" rule when you use BLASTP to search each protein from species A against all proteins from species B. A's best match to a B protein should then have that same A protein as its best match in the reciprocal search.

An alternative approach is not to rely on sequence similarity but functional similarity, as described recently. See my reply to a similar question here. This references a new paper by Nehrt and Hahn in which the authors look at comparative functional genomic data from mammals. It is true that functional and sequence similarity often overlap, but not always.

You might want to have a look at the so-called "reciprocal FASTA" method that is widely used to identify orthologs in different species. There are many papers using that technique where usually hits are identified as amino acid identity >40% over at least 80% of the total length of the protein...

I think the ACT (artemis tool (also web) of Sanger can do it as well using that method).

You might want to have a look at this recent (June 2011) review about orthology prediction methods:

http://www.ncbi.nlm.nih.gov/pubmed/21690100