The paper gives details of statistical modelling techniques which can be used to estimate risk and office premiums from past claims data. The methods described allow premiums to be estimated for any combinaton of rating factors, and produce standard errors of the risk premium. The statistical package GLIM is used for analysing claims experience, and GLIM terminology is used and explained thoughout the paper.Arguments are put forward for modelling frequency and severity separately for different claim types. Pitted values can be used to estimate risk premiums, and the incorporation of expenses allows for the estimation of office premiums. Particular attention is given to the treatment of no claim discount.The paper also discusses possible uses of the modelled premiums. These include the construction of 'standardised' one way tables and the analysis of experience by postal code and model of vehicle. Also discussed is the possibility of using the results for assessing the impact of competition, and for finding 'niche' markets in which an insurer can operate both competitively and profitably.

No abstract