We have studied how small changes in the distance between the TATA-box and cap site affect transcription of a eukaryotic gene in vitro and in vivo. The trout protamine gene TPG-3 [Gregory et al. (1982) Nucl. Acids Res. 10, 7581-7592] is a good model for such a study as it has (i) a consensus TATA-box 32 base pairs (bp) upstream from an A-residue which is the natural cap site (designated +1) (ii) two further A-residues at -5 and +5, providing alternative transcriptional start sites which are in significantly different sequence environments and (iii) a unique AvaII restriction site immediately downstream from the TATA-box which is ideal for the insertion or deletion of up to 3bp. Transcripts of the wild type and mutant genes were generated in vitro using a HeLa whole cell extract or 'in vivo' by transient expression following their transfection into HeLa cells. These 'spacer' mutations did not affect the efficiency of transcription of the gene in vitro but they did affect the selection of transcriptional start site both in vitro and 'in vivo'. Analysis of 5'-ends by S1-mapping and primer extension showed that the A-residue(s) selected are those which, by insertion or deletion, come to lie on the same face of the DNA double helix as the TATA-box, although the DNA sequence in the immediate vicinity of the potential start sites influences their utilisation. Comparison of the TPG-3 wild type transcripts in these experimental systems with natural mRNA suggests that cap site selection is more stringent in the developing trout testis.