First principal calculations made with the "Ether" code appeared pessimistic when compared with results for batteries with a duration from 50 s upwards. Subsequent examination into the reason for this confirmed that "Ether" predictions were correct in themselves, however a significant self-heating effect was necessarily taking place, this being reinforced by treatment of all the available information. The challenge for improved thermal battery modeling was then to determinate pertinent laws for this effect and to obtain accurate measurements of the phenomenon itself. An experimental method was used where the results obtained from reusable clamp stacks were subjected to mathematical treatment using "Ether". This has been successfully carried out in two similar cases and has clearly confirmed the premises: in these cases, the addition of a constant 400 W (kg of cells)(-1) internal heat generation allows for a close matching between numerical and experimental results globally, up to the exhaustion of the electrochemically active products. Correlative on the electrochemical side, when compared to single-cell tests, extended self-discharge clearly takes place in clamp stacks. Thermal batteries modeling - which has already been very useful for years - progresses toward best and appears on the way to make accurate predictions through deep physical knowledge, although there is still significant work to do. (c) 2004 Elsevier B.V. All rights reserved.