The presence and severity of compulsive behaviours may be evaluated via the Compulsive Behaviour Checklist (CBC) and this instrument has been successfully employed in people with intellectual disability. However, the applicability of the overall CBC scoring system, which entails tallying the number of behavioural categories represented (i.e. five) as well as the number of individual behaviours endorsed (i.e. 25), is not known in the population with Prader-Willi syndrome (PWS). The present investigation examined the latent variable structure of the CBC in people with PWS in order to identify possible population-specific scoring and interpretation considerations. The 25 behaviour-specific items of the CBC were analysed for 75 people with PWS (44 females and 31 males) aged between 4 and 41 years (mean +/- SD = 11.4+/-9.4) via factor analysis with principal component extraction and equamax rotation. The most suitable solution was determined on the basis of multiple empirical criteria: (1) the scree test; (2) eigenvalues >1.00; (3) salient loadings >0.30; (4) the clarity of item assignment to a single latent dimension; (5) the internal consistency of the latent dimension(s) (coefficient alpha > or = 0.70); and (6) item-total correlations between 0.20 and 0.79. In addition, solutions were examined with respect to psychological theory and previous research. A 'general factor' (i.e. single latent dimension) solution which adhered to all a priori criteria was indicated. Twenty-four out of 25 items achieved salient loadings ranging from 0.46 to 0.80 on the general factor. The single item which failed to achieve salience, 'deviant grooming-skin picking', exhibited both substantial unique variance (0.997) and moderate reliability (r = 0.59, P<0.001). The internal consistency of the general factor was strong (alpha = 0.93) and all salient items were suitably correlated with the unit-weighted total score (r(item-total) = 0.41-0.77). The traditional CBC scoring system, which includes tallying the number of categories represented, would not be relevant in this PWS sample. In addition, the recommended tallying of the number of individual behaviours endorsed does not reflect the empirically indicated notion of compulsive behaviour in this special population. These findings indicate that the 24 salient items should be scored as a unit-weighted composite and that the score on the substantially unique item (skin picking) should be considered a separate measure when evaluating compulsive behaviours via the CBC in people with PWS.