Previous studies have tentatively suggested that working memory training (WMT) has the potential to improve reward processing, but it is not known how long this improvement lasts, whether there is a lag effect, or whether it is reflected in neurophysiological indicators. In this study, 40 university students with subsyndromal depression were randomly assigned to a training group or a control group and completed a 20-day working memory training task and a simple memory task, respectively. All participants completed the Temporal Experience of Pleasure Scale (TEPS) and a doors task with electroencephalogram (EEG) signals recorded simultaneously on a pre- and post-test and a 3-month follow-up. The reward-related positivity (RewP) amplitude, theta power, and their differences between conditions (i.e., ΔRewP and Δtheta power, respectively) in the doors task were the primary outcomes, and the score on TEPS was the secondary outcome. The results indicated no group-related effects were demonstrated in primary and secondary outcomes at post-test and 3-month follow-up. Furthermore, the differences in the pre- and post-test in Δtheta power were moderated by the baseline severity of depression. This was primarily driven by the fact that the change values in the control group increased with the severity of depression, while the change values in the training group had high homogeneity. Our findings did not provide support for the effect of WMT on reward processing across the whole sample, but without intervention, there would be high heterogeneity in the change in the cognitive control ability to loss feedback, which is detrimental to individuals with high depression severity.