Computational approaches to reinforcement learning (RL) suggest that RL happens model-free (MF) via reward prediction errors and model-based (MB) via state prediction errors not only during instrumental (Daw et al., 2011), but also during Pavlovian learning (Schad et al., 2020). Pavlovian-instrumental transfer (PIT) reflects the influence of Pavlovian values on instrumental responses. In single-lever PIT paradigms, which assess the rate of a single instrumental response (Garbusow et al., 2014), PIT effects are often thought to be only MF (Review Cartoni et al. 2016), and they are found to correlate with reduced MB instrumental control (Sebold et al., 2016). However, whether single-lever PIT effects are always MF, or can also be MB, is unclear. Here, we developed a novel two-stage paradigm, designed to assess the contribution of MF and MB control in a single-lever PIT task. While prior tasks have assessed the influences of Pavlovian learning on PIT across blocks of trials, our task assesses this at a trial-by-trial level. We simulated data with a computational dual-control model, assuming separate MF versus MB Pavlovian learning systems. The simulations showed that a two-way interaction (outcome x CS-equality, i.e., whether in a trial the CS was the same in Pavlovian learning and PIT) indicated MF PIT, while a three-way interaction (outcome x CS-equality x transition probability) indicated MB PIT. Preliminary data shows that half of the subjects qualitatively showed MF and MB influences on PIT, in line with Pavlovian learning, and single-lever PIT tasks specifically, involving not only MF but also MB computations.