Humans are prone to decision biases, which make behavior seemingly irrational. An important cause for decision biases is that the context in which decisions are made can later influence which choices humans prefer in new situations. Current computational models (e.g. relative value learning or range normalization) often require extensive environmental knowledge to explain these biases. Here, we tested the hypothesis that decision biases are mainly driven by a tendency to repeat context-specific actions. We implemented a series of nine value-based decision-making tasks on n = 351 male and female participants and reanalyzed six previously published datasets (n = 350 participants). We found that higher within-context repetition of an option was associated with biased choices including higher subjective valuation and lower uncertainty for repeated actions. Next, we used a hierarchical Bayesian reinforcement learning model based on two basic principles, learning by reward and action repetition and tested it on all datasets. Our results show that the combination of these two basic principles is sufficient to explain biased choices in stable environments. We demonstrate via extensive model comparison that our model outperforms all tested alternatives (implementations of value normalization and a goal centric account). These results provide insights into decision biases during value-based decision-making and suggest a parsimonious mechanism for understanding habit-like choice tendencies.