Visual working memory (VWM) is limited in both the capacity of information it can retain and the rate at which it encodes that information. We examined the influence of stimulus complexity on these 2 limitations of VWM. Observers performed a change-detection task with English letters of various fonts or letters from unfamiliar alphabets. Average perimetric complexity (κ)—an objective correlate of the number of features comprising each letter—differed among the fonts and alphabets. Varying the time between the memory array and mask, we used change-detection performance to estimate the number of items held in VWM (K) as a function of encoding time. For all alphabets, K increased over 270 ms (indicating the rate of encoding) before reaching an asymptote (indicating capacity). We found that rate and capacity for each alphabet were unrelated to complexity: Performance was best modeled by assuming that both were limited by number of items (K), rather than by number of features (K × κ). We also found a higher encoding rate and capacity for familiar alphabets (∼45 items s−1; ∼4 items) than for unfamiliar alphabets (∼12 items s−1; ∼1.5 items). We then compared the familiar English alphabet to an unfamiliar artificial character set matched in complexity. Again, rate and capacity was higher for the familiar than for the unfamiliar stimuli. We conclude that rate and capacity for encoding into visual working memory is determined by the number of familiar feature-integrated object representations.