Humans show remarkable flexibility in adapting their behaviour to constantly changing environments. This flexibility relies on the ability to regulate motivation in response to changing motivational demands. Typically, the amount of effort required to achieve a certain goal is not precisely signalled by the environment but needs to be learnt from experience. By contrast, prior work examining motivated choices has usually directly instructed effort requirements. It therefore remains unclear how healthy individuals estimate and flexibly regulate effort and how they might achieve this in dynamically changing environments. In the current study, we examine how effort learning is shaped by different types of environmental uncertainty when motivational requirements are not explicitly instructed. Analogous to tasks in the reward learning domain, we designed a novel effort learning task that systematically manipulated two key sources of uncertainty: volatility and noise. Participants were asked to exert effort by squeezing hand-held dynamometers. We characterised effort learning across different stages of the effort production process (e.g., initiation of effort production, effort expectation, error-driven adjustment), which allowed us to capture the dynamics underlying effort estimation and regulation over time. Our findings reveal that humans are able to learn effort requirements by integrating both effort priors and sensorimotor feedback. We further show that effort learning is modulated by environmental statistics, with slower force initiation, weaker priors, slower learning, and faster within-trial force adjustments in high noise environments, but slower learning and slower within-trial force adjustments in high volatility environments. In summary, when effort information is not instructed, different sources of uncertainty about an action's required effort are integrated into participants' internal priors to flexibly guide effort exertion. Our work may provide a useful framework for understanding motivational disorders where abnormal effort learning and estimation may underlie the reduced willingness to exert effort for reward.