The cognitive model of problem gambling posits that erroneous gambling-related fallacies are key in the development and maintenance of problem gambling. However, this contention is based on cross-sectional rather than longitudinal associations between these constructs, and gambling fallacy instruments that may have inflated this associated by their inclusion of problem gambling symptomatology. The current research re-evaluates the relationship between problem gambling and gambling-specific erroneous cognitions in a 5-year longitudinal study of gambling using a psychometrically sound measure of erroneous gambling-related cognitions. The sample used in this study (n = 4,121) was recruited from the general population in Ontario, Canada, and the retention rate over 5 years was exceptionally high (93.9%). The total sample was similar, in age and gender distributions, to the census data at the time of data collection for Canadian adults (18–24 years, n = 265, 55.8% female; 25–44 years, n = 1,667, 56.4% female; 45–64 years, n = 1,731, 55.4% female; 65 + years, n = 458, 44.75% female). Results of both cross-sectional and longitudinal analyses confirm that gambling-specific fallacies appear to be etiologically related to the subsequent appearance of problem gambling, but to a weaker degree than previously presumed, and in a bidirectional manner.