Large-scale sequencing of a 3'-cDNA library permits one to analyse gene expression profiles in various tissues. However, many such sequences lack enough information about the encoded proteins. To overcome this problem, we tested a new library, consisting of a 3'-directed cDNA sequence fused to a to a 5' sequence of about 300 bp. Such 'joint molecules' of about 600 bp were amplified by PCR and directly sequenced. About 40% of these joint molecules included the 5' and 3' terminal portions of the mRNA, and most of the remaining clones contained the middle portion and 3' end of the mRNA. The upstream sequences contained sufficient information with which to search for similarity, ORFs, motifs and hydropathy, thus allowing the mRNAs to be categorized and their functions predicted. The rapid categorization of the cDNAs will help to sort those clones that merit further analysis.