From picking dinner to college courses, the choices we make reflect our past experiences and momentary preferences. The value of a choice is inherently subjective, internal to a decision maker, yet we need to identify objective measures to study its neural basis. According to learning theory, value can be inferred from reinforcement history through an algorithmic computation (‘model-inferred value’). In contrast, in behavioral economics, appropriately designed tasks can elicit choice patterns that reveal subjective preferences (‘revealed choice value’). However, the relationship between these distinct measures of subjective value remains unclear, which limits studying its neural basis. We designed a probabilistic reward learning task that brought together these two approaches. Rats chose between two options that were probabilistically ‘baited’ with rewards, varying across blocks. To earn a reward, rats committed to their choice by investing time for uncertain, delayed rewards. Choice value could be inferred from reward and choice history using generalized linear models or fitting reinforcement learning models. These model-inferred choice values robustly predicted rats’ choice behavior, as well as the magnitude of time investment. In turn, shorter time investments predicted choosing the lower-valued (‘model-inconsistent’), rather than the higher-valued (‘model-consistent’) option. Thus, time investment reflected the rats’ subjective choice valuation, beyond model-inferred value alone, thereby behaviorally revealing choice value. To investigate the neural processes underlying choice valuation, we monitored dopamine release in the ventral striatum using fiber photometry with virally expressed, genetic dopamine sensors. We observed phasic dopamine release at the time of choice, which strongly predicted trial-by-trial time investment seconds in advance, but was not correlated with model-inferred choice value based on reward history. Thus, mesolimbic dopamine encodes the subjective valuation of choice options that can be behaviorally read-out in single-trial time investment decisions.