Identifying which variables modify a treatment effect, and thus determining for whom an intervention works, is a central goal in clinical and applied psychology. In practice, this boils down to identifying causal treatment effect modifiers: the specific variables driving individual differences in treatment efficacy. Randomized controlled trials are often unfit for this, as they typically randomly assign the treatment, but not the candidate modifiers. Alternative causal discovery methods have restrictively strong assumptions that limit their usefulness in practical settings. To bridge this gap, we propose a novel causal discovery method utilizing effect-invariance. By comparing datasets from multiple environments, it exploits shifts in the covariate distribution to isolate causal effect modifiers. We show the method’s strict false-positive bounds and assess finite-sample performance via simulations.