Spliceosomes are assembled in stages. The first stage forms complex E, which is characterized by the presence of U1 snRNPs base-paired to the 5' splice site, components recognizing the 3' splice site and proteins thought to connect them. The splice sites are held in close proximity and the pre-mRNA is committed to splicing. Despite this, the sites for splicing appear not to be fixed until the next complex (A) forms. We have investigated the reasons why 5' splice sites are not fixed in complex E, using single molecule methods to determine the stoichiometry of U1 snRNPs bound to pre-mRNA with one or two strong 5' splice sites. In complex E most transcripts with two alternative 5' splice sites were bound by two U1 snRNPs. However, the surplus U1 snRNPs were displaced during complex A formation in an ATP-dependent process requiring an intact 3' splice site. This process leaves only one U1 snRNP per complex A, regardless of the number of potential sites. We propose a mechanism for selection of the 5' splice site. Our results show that constitutive splicing components need not be present in a fixed stoichiometry in a splicing complex.