We link Friedman’s partial dependence plot with Pearl’s backdoor adjustment formula. We discuss situations when possible causal interpretations can be made for black-box machine learning models.