Model pre-mRNAs containing two introns and three exons, derived from the human beta-globin gene, were used to study the effects of internal exon length on splice site selection. Splicing was assayed in vitro in HeLa nuclear extracts and in vivo during transient expression in transfected HeLa cells. For substrates with internal exons 87, 104, and 171 nucleotides in length, in vitro splicing proceeded via a regular splicing pathway, in which all three exons were included in the spliced product. Primary transcripts with internal exons containing 23, 29, and 33 nucleotides were spliced by an alternative pathway, in which the first exon was joined directly to the third one. The internal exon was missing from the spliced product and together with two flanking introns was included in a large lariat structure. The same patterns of splicing were retained when transcripts containing 171-, 33-, and 29-nucleotide-long internal exons were spliced in vivo. A transcript containing a 51-nucleotide-long exon was spliced in vitro via both pathways but in vivo generated only a correctly spliced product. Skipping of short internal exons was reversed both in vitro and in vivo when purines in the upstream polypyrimidine tract were replaced by pyrimidines. The changes in the polypyrimidine tract achieved by these substitutions led in vitro to complete (transcripts containing 28 pyrimidines in a row) or partial (transcripts containing 15 pyrimidines in a row) restoration of a regular splicing pathway. Splicing in vivo of these transcripts led exclusively to the spliced product containing all three exons. These results suggest that a balance between the length of the uninterrupted polypyrimidine tract and the length of the exon is an important determinant of the relative strength of the splice sites, ensuring correct splicing patterns of multiintron pre-mRNAs.