Memory comes in different varieties that each have their characteristic properties, including working memory (WM), short-term memory (STM) and long-term memory (LTM). LTM is what we store for more than, say, a minute, and has essentially unbounded storage capacity. STM is used at shorter retention intervals, but its capacity is unclear, though often assumed to be limited. WM is used to temporarily retain items that are manipulated by ongoing cognitive operations, and is thought to have a capacity of 3 or 4 items. Here, I show that the evidence for such severely capacity-limited memory stores mostly comes from experiments with substantial interference among items, and that such capacity limits are removed once interference among items is reduced. Further, I prove mathematically that, under general conditions, the presence of interference among items guarantees fixed and limited capacity limits. I also present experiments suggesting that the forms of attention that supposedly yield capacity limitations of 3 or 4 items have fundamentally different properties from memory, suggesting that they cannot be the reason for WM limitations. Based on these and other experiments, I propose a unitary view of STM, LTM and WM. With brief presentation durations, memory has a large capacity but is fleeting. When memory items are presented repeatedly or for longer durations, they become gradually stabilized into LTM representations. If proactive interference is added, memory capacity becomes compatible with past WM capacity estimates.