Abstract: Persistent activity has been interpreted as the neural substrate underlying working memory for decades. However, recent studies suggested that these persistent activity findings might be an “artifact” of trial-averaging. In single-trial analyses, activity often occurs in sparse, synchronous bursts, both for single neurons and local networks. Alternative working memory models depending on the short-term synaptic plasticity and rhythmicity of discharges were examined and concluded that each of these models could explain a range of memory-guided behaviors which are hard to be explained with persistent activity models. Some attempts to reconcile the discrepancy of persistent activity and dynamic coding frameworks were introduced. In the end, I pointed out some directions which are worth considering in the future.