The candy-stripe spider, Enoplognatha ovata, exhibits a striking color polymorphism comprising three morphs. A number of lines of evidence strongly suggest that this polymorphism is maintained by natural selection: its presence in a sister species, E. latimana; the physical nature of the variation; the virtual lack of monomorphic populations; the highly consistent rank-order of morphs within populations; and the presence of large-scale clines associated with climatic variables. However, the absence of selection is equally strongly suggested by very local surveys of morph frequencies over space and time, perturbation experiments, and a variance in morph frequency between populations that is virtually independent of spatial scale. In addition, local spatial patterns in one study site (Nidderdale, Yorkshire, England) have been explained in terms of intermittent drift over half a century ago, a hypothesis supported here by the distributions of four other genetic markers (two allozyme and two visible polymorphisms). A heuristic model is suggested that reconciles these apparently contradictory messages regarding the importance of drift and selection in this system. It is proposed that when allele frequencies of the color morph redimita lie between approximately 0.05 and 0.3, the Δq on q plot is very shallow, so that within this region, where the majority of populations lie, selection is weak and drift is the major force determining local morph frequencies. However, outside this range of frequencies, powerful selection acts to protect the polymorphism. This model may apply to polymorphisms in other species and explain why evidence of selection in natural populations is often elusive.