A deterministic view of price indices offers no measure of index reliability, a shortcoming that can be overcome by using regression analysis. Recognising that Seemingly Unrelated Regressions estimation (rather than simple regression analysis) may be appropriate, more precise interval estimates of price indices can be computed. We show that the regression approach to price index construction has benefits beyond those previously discussed in the literature. Within this framework one can determine the appropriate estimation technique and test for significant changes in the price level. The statistical sensitivity of an index to the composition of the underlying "basket" of goods can also be tested.