The field of policy learning is characterised by a proliferation of concepts and lack of systematic findings. However, the literature has struggled to move beyond the seminal contributions made more than three decades ago however. We argue that different strands in the literature have failed to communicate because of selection bias and ambiguity about what constitutes evidence of learning in public policy. This conceptual paper suggests a typology wide enough to capture the variation observed by different strands of literature on learning in public policy. We first present a basic model of policy learning based on four categories. The model arises out of an overview of political science studies on learning. We then generate variation within each cell by using rigorous concepts drawn from behavioural research. Specifically, we conceptualize learning as control over the contents and goals of knowledge. By looking at learning through the lenses of knowledge utilization, we show that the basic model can be expanded to reveal sixteen types of learning. These sixteen types are all well-established in the literature, but up until now the scope conditions and connections among types have not been clarified. By providing a comprehensive typology, we aim to lay the foundations for the systematic comparison across and within cases.