Although measures of actual and perceived physical ability appear to predict falls in older adults, a disparity between these two, also known as misjudgement, may even better explain why some older adults fall, while their peers with similar abilities do not. Therefore, we investigated whether adding a misjudgement term improved prediction of future falls. Besides conventional measures of actual (physical measures) and perceived abilities (questionnaires), we used a stepping down paradigm to quantify behavioural misjudgement. In a sample of 55 older adults (mean age 74.5 (s.d. = 6.6) years, 33 females and 20 fallers over a 10-month follow-up period), we tested the added value of a misjudgement term and of a stepping-down task by comparing experimental Bayesian logistic-regression models to a default null model, which was composed of the conventional measures: Falls Efficacy Scale international and QuickScreen. Our results showed that the default null model fitted the data most accurately; however, the accuracy of all models was low (area under the receiver operating characteristic curve (ROC) ≤ 0.65). This indicates that neither a misjudgement term based on conventional measures, nor on behavioural measures improved the prediction of future falls in older adults (Bayes Factor ₁₀ ≤ 0.5).