This paper addresses the question of whether there is an alternative to asking a direct religious question in the 2001 Census through a systematic review of alternative methods of quantifying religious populations. After establishing the limited availability of direct information on religion, the paper considers the application of an inferential method for estimating religious populations based on data from the 1994 Fourth National Survey of Ethnic Minorities (NSEM). Except for the monoreligious Pakistani and Bangladeshi populations, evidence of substantial spatial variation in the association between ethnicity and religion severely limits application of the method below national level. Using the religiously diverse Indian population of outer London as a case study, the NSEM is investigated to see whether the incorporation of other ¿predictors¿ of religion as weighting variables improves sensitivity of the method to this variation. In a second method, logistic regression is employed to develop predictive models of religious affiliation for application to census microdata. Based on these two applications, it was concluded that very little of the variation in the religious geography of Indians in outer London can be inferred from measured characteristics in the census. However, this does not necessarily imply support for inclusion of a religion question in the 2001 Census. It was suggested that it may be more appropriate and cost effective to define and solve the need for data locally.