Given that interesting HIV vaccine candidates, including live preparations and DNA plasmids, exist and that the first phase III vaccine (AIDSVAX) testing is due to begin this summer, 1998, in the U.S., adequately addressing trial preparedness is a pressing issue. Despite double-blind randomized controlled clinical trial design, there may be difficulties with interpretation and use of the usual measures of vaccinal efficacy and calculation of sample size. Difficulties arise from vaccine characteristics (e.g. mode of action, time-lag, waning) and population heterogeneities (e.g. differences in susceptibility, sexual behaviour, mixing preferences) causing frailty effects that can exacerbate bias and time-dependent effects already known to exist in simple cases. Since vaccine properties, particularly mode of action, are unlikely to be known before the onset of clinical trials, choosing an efficacy measure and the associated analyses and sample size calculations will be problematic. Interim analyses designed to decide whether a study will be prolonged may be tenuous if based on a time-dependent measure and will influence sample size determination. Despite shortcomings, general recommendations can be made to minimise pernicious effects. The objectives of this paper are principally to review the current state of knowledge of the different stages in the preparation of large phase III HIV vaccine efficacy trials, the methodological difficulties related to their design, and the analysis of data collected from them. Mathematical models and trial simulations are used to demonstrate that further research is necessary to study the behaviour of vaccine efficacy measures under heterogeneous conditions of population, vaccine action, and trial design and identify a time-independent efficacy measure. Alternative methods to validate sample size calculations have to be developed in older to reduce the chances of unnecessary economic and human cost in phase III HIV vaccine trials.