INTRODUCTION Findings regarding the association between overweight and all-cause mortality range from significantly lower to higher risk, compared with body-mass-index (BMI) within the "normal" range. METHODS We examined empirically potential methodological explanations for these apparently conflicting results using questionnaire and linked mortality data from 246,314 individuals aged ≥45 years in the Australian 45 and Up Study (11,127 deaths; median follow-up 3.9 years). Hazard ratios (HR) for all-cause mortality associated with BMI were modelled according to different methods of accounting for illness at baseline, finer versus broader gradations of BMI and choice of reference group, adjusting for potential confounders. RESULTS In analyses using the broad World Health Organization (WHO) categories, the all-cause mortality HR was significantly lower in the overweight category (25.0-29.99 kg/m²), than the normal weight (18.5-24.99 kg/m²) category. However, in analyses accounting for baseline illness, which excluded those with pre-existing illness at baseline, ever-smokers and the first 2 years of follow up, absolute age-standardised mortality rates varied up to two-fold between finer BMI categories within the WHO normal weight category; rates were lowest at 22.5-24.99 kg/m² and mortality HRs increased steadily for BMI above (p(trend)<0.02) and below (p(trend)<0.003) this reference category. Hence, the breadth of the BMI categories used and whether or not baseline illness is accounted for explain the apparent discrepancies between reported BMI-mortality associations. CONCLUSION Using fine BMI categories and the category with the lowest absolute rates as the reference group and accounting for the potential confounding effects of baseline illness is likely to yield the most reliable risk estimates for establishing the independent relationship of BMI to all-cause mortality. These results and those of other studies indicate that a BMI of 22.5-24.99 kg/m², not the broad "overweight" category of 25-29.99 kg/m², was associated with the most favourable mortality risk.