A major goal of cognitive science is to characterize how an individual's past experiences guide their present decisions in a sequential task. Various empirical evidence support a process of incremental learning, well-characterized by the framework of reinforcement learning, whereby repeated exposures to similar situations shape decisions. However, in a complex world with sparse data a more sample-efficient process is needed. Prior work has suggested that episodic memory supports decision-making in such settings. Here, we provide novel behavioral evidence that episodic memory supports decision-making in temporally extended settings. We propose that value-based decision-making and episodic memory share common mechanisms to encode and retrieve past events, which in turn shape option evaluation and ultimately choice. In two experiments, we empirically test hypotheses that relate classic dynamics of sequential episodic memory retrieval to response patterns in novel evaluation and decision tasks. We find subjects' reported value estimates are subject to biases analogous to classic episodic memory biases (Experiment 1), and their choices are best captured by an episodic recall-based model (Experiment 2). These results suggest a novel link between value-based decision-making and episodic memory, which could reflect a psychologically plausible mechanism for computing decision variables by Monte Carlo sampling.