Martempering is well established as a method by which the stresses and strains generated during the quenching of a steel component may be controlled. In this process the component is held at an intermediate temperature in a salt bath subsequent to the initial cooling from the austenitization temperature, but prior to the final quench to ambient. This allows sufficient time for most of the temperature gradient in the component to be eliminated before the transformation of the austenite to martensite in the final stages of the cooling process. It is essential that the intermediate treatment does not cause decomposition of the austenite by diffusional processes, to form soft transformation products. This limits the time available for the intermediate treatment. Doubts have been raised about the fundamental assumptions on which the process is based, since the absence of a vapour blanket stage in the initial cool in the salt bath produces very high surface heat transfer coefficients at the start of the process, when the material is very soft. It has been the purpose of the present investigation to verify these assumptions by the use of a mathematical model of the stress generation process. In the case of 835M30 steel, 87 seconds at 450°C reduced the temperature gradient in a 15 mm thick plate to 5°C, without producing any diffusional decomposition to soft structures. However, the beneficial effect of this small temperature gradient during the later stages of the process, was negated as far as strain was concerned by the increased distortion that occurred in the very early part of the cooling process in the salt bath. This latter effect was due to the very large surface heat transfer coefficient provided by the quenchant at this stage. The martempering operation, however, did successfully remove the thermal stresses by the time the component reached ambient. Modification of the process to include an oil quench from the salt bath was not beneficial, as the stresses and strains produced at the end of the quench were very similar to those produced by a conventional oil quench from the austenitization temperature. The calculated residual stress and strain distributions were checked against the corresponding experimental values. Agreement was reasonably good, except for a small but uniform discrepancy in the case of the strains.