Popular models of visual working memory assume that working memory is limited by a constant resource, which is conceived as either quantized (as in slot models) or infinitely divisible. These models share the assumption that the probability and quality of retrieval depends on the resource assigned to a representation in working memory. We present an alternative model that incorporates the principles of general theories of memory: Retrieval is cue-based, and performance is limited by interference arising from several sources. Items compete for retrieval according to the amount of evidence from memory in favor of each item. Evidence arises from three sources: Persistent activation of item representations, activation from cueing at retrieval, and background noise. One item is held in the focus of attention; this item is represented with higher precision, and suffers less interference from competing items. We competitively test the interference model against the slot-averaging model (Zhang & Luck, 2008), the resource model (Bays et al., 2008) and the Variable Precision model (Van den Berg et al., 2012), using data from three experiments using the color-reproduction paradigm. Experiment 1 varied memory set size and presentation duration; Experiments 2 and 3 investigate the effects of a pre-cue and a post-cue, respectively.