Another one, the ChiP mapper2 http://genome.ufl.edu/mapper/mapper-main, which uses the TRANSFAC and JASPAR databases in combination with other algorithms.
There's also a really easy to use COTRASIF tool http://biomed.org.ua/COTRASIF/pwmfinder/start.html , which can search for JASPAR/TRANSFAC7/custom matrices in the gene-upstream regions of a fixed number of genomes, and can look for binding sites conserved across species in the upstream regions of orthologous genes. The only problem with that tool is that it's databases (promoters, JASPAR, and orthology) haven't been updated for about two years. (Disclosure: I'm the developer of COTRASIF, and this question made me schedule a long-required databases update for this weekend :) )
@Simon Rasmussen: There might be some redundancies in the matrix library of the TRANSFAC database, but that TRANSFAC is not curated is certainly wrong. There is a considerable staff doing nothing else but curating TRANSFAC.
To predict TFBSs on DNA sequences, sure you are interested to use TRANSFAC database. It contains manually curated collection of positional weight matrices (PWMs). There are matrices of several origins, including at least those several groups: (1) collected from publications, (2) constructed by the team based on careful collection of individual binding sites experimentally proven and published, (3) constructed by the team based on published ChIP-chip and ChIP-seq experiments. Because of the policy to collect all published matrices, there are sometimes two or more matrices for the same factor, which probably is considered as redundancy. Please note that matrices for the same factor published by different groups and based on different experiments are not exactly the same, and they can result in different matches.
For those who prefer to use a collection like "one matrix per transcription factor",
TRANSFAC provides non-redundant profile, where the best matrix for each TF is pre-selected by the team based on a number of criteria to ensure matrix quality.