In the literature on working memory (WM), a paradox exists according to which very similar memory tasks provide support for very different estimates of working memory capacity. The current paper analyses the conflicting estimates of a capacity of 4+/-1 with a capacity of 1. To this end a dynamic process model of short-term recognition is used to generate data to which exponential speed-accuracy trade-off functions are fitted. The results show that even though the process model has a capacity larger than one, the exponential SAT functions indicate a one-chunk hypothesis. Further nested modeling reveals, counter to the dominant belief, that retrieval rate is insensitive to differences in WM capacity. The resolution of the WM capacity paradox lies in the choice of dependent measure.