The pressure and temperature conditions of formation of natural diamond can be estimated by measuring the residual stress that an inclusion remains under within a diamond. Raman spectroscopy has been the most commonly used technique for determining this stress by utilising pressure-sensitive peak shifts in the Raman spectrum of both the inclusion and the diamond host. Here, we present a new approach to measure the residual stress using quantitative analysis of the birefringence induced in the diamond. As the analysis of stress-induced birefringence is very different from that of normal birefringence, an analytical model is developed that relates the spherical inclusion size, R i, host diamond thickness, L, and measured value of birefringence at the edge of the inclusion, Updelta n(R exti ) extav Unknown control sequence 'Updelta', to the peak value of birefringence that has been encountered; to first order Updelta n extpk = (3/4)(L/R exti ) Updelta n(R exti ) extav Unknown control sequence 'Updelta'. From this birefringence, the remnant pressure (P i) can be calculated using the photoelastic relationship Updelta n extpk = - (3/4)n3 q extiso P exti Unknown control sequence 'Updelta', where q iso is a piezo-optical coefficient, which can be assumed to be independent of crystallographic orientation, and n is the refractive index of the diamond. This model has been used in combination with quantitative birefringence analysis with a MetriPol system and compared to the results from both Raman point and 2D mapping analysis for a garnet inclusion in a diamond from the Udachnaya mine (Russia) and coesite inclusions in a diamond from the Finsch mine (South Africa). The birefringence model and analysis gave a remnant pressure of 0.53 ± 0.01 GPa for the garnet inclusion, from which a source pressure was calculated as 5.7 GPa at 1,175°C (temperature obtained from IR analysis of the diamond host). The Raman techniques could not be applied quantitatively to this sample to support the birefringence model; they were, however, applied to the largest coesite inclusion in the Finsch sample. The remnant pressure values obtained were 2.5 ± 0.1 GPa (birefringence), 2.5 ± 0.3 GPa (2D Raman map), and 2.5–2.6 GPa (Raman point analysis from all four inclusions). However, although the remnant pressures from the three methods were self-consistent, they led to anomalously low source pressure of 2.9 GPa at 1,150°C (temperature obtained from IR analysis) raising serious concerns about the use of the coesite-in-diamond geobarometer. An erratum for this article appeared in Contributions to mineralogy and petrology, Vol. 162, No. 5, and may be found at http://dx.doi.org/10.1007/s00410-011-0694-4.