Haavelmo’s (1944) monograph “The Probability Approach in Econometrics” constitutes the probabilistic foundation for modern econometrics. It is based on a strict adoption of Neyman-Pearson (NP) testing theory, in which probability is defined as a long-run relative frequency, and where statistical inference is thought of as deductive reasoning with focus on long-run error control in repeated applications. NP theory does not accommodate inductive probabilistic inference from a unique set of data to an underlying population or theory model, which is precisely the main aim of Haavelmo. Thus, I argue that Haavelmo’s adoption of NP theory is inconsistent. In his attempt to make frequentist hypothesis testing fit his purpose, Haavelmo resorted to the notion of a hypothetical infinite population and a distinctly non-frequentist (Bayesian-like) interpretation of probability, in direct contradiction with NP theory. I also discuss the legacy of Haavelmo’s probability approach, and how it has permeated econometric theory and practice.