This paper models hot compression testing using a dilatometer in loading mode. These small-scale tests provide a high throughput at low cost, but are susceptible to inhomogeneity due to friction and temperature gradients. A novel method is presented for correcting the true stress-strain constitutive response over the full range of temperatures, strain-rates and strain. The nominal response from the tests is used to predict the offset in the stress-strain curves due to inhomogeneity, and this stress offset Δσ is applied piecewise to the data, correcting the constitutive response in one iteration. A key new feature is the smoothing and fitting of the flow stress data as a function of temperature and strain-rate, at multiple discrete strains. The corrected model then provides quantitative prediction of the spatial and temporal variation in strain-rate and strain throughout the sample, needed to correlate the local deformation conditions with the microstructure and texture evolution. The study uses a detailed series of 144 hot compression tests of a Zr-Nb alloy. While this is an important wrought nuclear alloy in its own right, it also serves here as a test case for modelling the dilatometer for hot testing of high temperature alloys, particularly those with dual α-β phase microstructures (such as titanium alloys).