Abstract In this paper we employ explainable artificial intelligence methods to identify unfairness in mortgage lending. Our aim is to reproduce credit lending decisions via explainable machine learning models and, then, assess whether such decisions are fair, particularly in terms of race. To this end, the paper employs data from New York state, deriving from the Home Mortgage Disclosure Act (HMDA). We contribute to the existing literature in two main ways. First, we assess fairness marginally, by means of parity measures based on the recently proposed S.A.F.E. AI metrics; but also conditionally, comparing the explanations in different population groups. Second, we extend the Shapley value approach measuring the contribution of each explanatory variable not to the predicted values but to precision and recall, thereby better taking into account data unbalancedness. Our empirical findings indicate the presence of racial disparities in loan approval rates. This underscores the need for increased efforts and targeted interventions to promote fair and equitable lending practices.