The Annett Hand Preference Questionnaire (AHPQ) was administered to a sample of 352 randomly selected individuals from the general community to examine the reproducibility of its handedness classification and to evaluate its model-based reliability and convergent validity. Latent class analysis showed that the eight categories of hand preferences could not be justified on statistical grounds. Instead, three broad handedness classes adequately accommodated the variety of handedness patterns: 'consistent right' (66.0%), 'consistent left' (9.8%) and 'inconsistent or mixed' (24.2%). Confirmatory factor analysis not only showed that the AHPQ is reliable and has solid convergent validity, but also the measurement properties of the AHPQ could be further improved by eliminating a few items from the scale. The implications of these findings are discussed, and it is suggested that questionnaire requires modification, possibly by replacing obsolete items such as 'sweeping' and 'shovelling' with modern manual activities, such as 'typing SMS messages' and 'using a remote control'.