Working memory is typically defined as a system devoted to the simultaneous maintenance and processing of information. However, the interplay between these 2 functions is still a matter of debate in the literature, with views ranging from complete independence to complete dependence. The time-based resource-sharing model assumes that a central bottleneck constrains the 2 functions to alternate in such a way that maintenance activities postpone concurrent processing, with each additional piece of information to be maintained resulting in an additional postponement. Using different kinds of memoranda, we examined in a series of 7 experiments the effect of increasing memory load on different processing tasks. The results reveal that, insofar as attention is needed for maintenance, processing times linearly increase at a rate of about 50 ms per verbal or visuospatial memory item, suggesting a very fast refresh rate in working memory. Our results also show an asymmetry between verbal and spatial information, in that spatial information can solely rely on attention for its maintenance while verbal information can also rely on a domain-specific maintenance mechanism independent from attention. The implications for the functioning of working memory are discussed, with a specific focus on how information is maintained in working memory.