We begin by considering the lack of methodology for obtaining second moments of outstanding claims in non-life insurance (sections 1 and 2), and give some arguments as to why this lack is to be deprecated (section 3). We suggest that part of the reason is the lack of generality of the models and methods currently in use in claims analysis, and we suggest further that some unification might be achieved through regression analysis (section 4). The claims analysis problem is formulated and solved in terms of regression methods (sections 5 to 8), estimates of both first and second moments of outstanding claims being obtained. Two specific examples of the recommended regression procedures are presented algebraically (section 9). One of these is extended to a numerical example (section 10). The paper concludes with some general observations (section 11).