Aim The present study aimed to explore the two-year naturalistic trajectory of time- and event-based prospective memory (PM) in patients with first-episode schizophrenia. Methods We administered a computer-based dual-task PM paradigm to 57 individuals with first-episode schizophrenia at baseline and after 6 months, 12 months and 24 months. Forty-eight healthy controls were also recruited and completed all the measures at baseline. We compared the trajectories between time-based and event-based PM in first-episode schizophrenia patients using repeated measures ANOVAs, and examined the relationship between PM and clinical symptoms using Spearman's correlation. Results PM impairments improved significantly after 24 months of follow-up. However, time-based and event-based PM appeared to run different trajectories. After 24 months, first-episode schizophrenia patient performed poorer than healthy controls in time-based but not event-based PM. PM did not appear to be correlated with clinical symptoms, both cross-sectionally and longitudinally. Conclusions This is one of the longest follow-up studies investigating PM in first-episode schizophrenia. Our results provide evidence to support that time-based PM is more temporally stable than event-based PM.