Polyploid evolution is often considered a mechanism of instant speciation; yet the establishment of rare tetraploids within diploid populations may be constrained by a frequency–dependent mating disadvantage (minority cytotype exclusion principle). I tested this hypothesis using experimental populations of Chamerion angustifolium (Onagraceae) that contained different proportions of tetraploids and diploids. Fitness, measured as total seed production over the entire flowering season, was calculated from a census of flower number and estimates of ovule number per flower and proportion of seed set per fruit. The fitness of tetraploids relative to diploids was frequency dependent, increasing from 0.4, when tetraploids were rare, to 0.7 when at 50% and 1.15 when they were in the majority (67%). This pattern exists because of a negative relationship between tetraploid frequency and seed set per fruit in diploids. Seed set in tetraploids was independent of cytotype frequency. The frequency–independent effect in tetraploids reflects higher assortative mating, partly because of non–random patterns of bee visitation. Bees visited a disproportionately high number of diploid inflorescences; however, the proportion of successive flights between tetraploids increased above random expectations as the frequency of tetraploids decreased. These results provide the first experimental test of frequency–dependent fitness in diploid–polyploid mixtures and suggest an important role for more gradual, population processes governing the evolution of polyploidy in natural populations.