Introduction Race has been widely used in studies on health and healthcare inequalities. Validity and reliability problems with race measurement are of concern in public health. This study investigated the agreement between self-assessment and observer's assessment of the skin colour classification, considering the influence of socioeconomic and demographic factors as well as food insecurity.

Methods A cross-sectional population-based study (1085 households) was carry out in 2005, in metropolitan area of Rio de Janeiro, Brazil. Food insecurity was estimated according Brazilian Scale of Food Insecurity. Race/skin colour was evaluated in five characteristics in according to Brazilian scale (white, black, mulatto, Asian and indigenous) by the interviewers and in an open-ended question answered by the respondent self-classified race spontaneously. The reliability between self- and interviewer-classification was estimated using agreement proportions and a weighted κ coefficient (k) for the full sample, stratified by socio-economic factors and food insecurity.

Results White individuals were classified most consistently, particularly those with more favourable socioeconomic conditions. Weighted κ values were higher at higher income levels (k=0.79; 95% CI 0.77 to 0.80) and higher levels of education (k=0.79; 95% CI 0.75 to 0.81). The most severe levels of food insecurity were strongly associated with low of socioeconomic strata. The consistency of classification of race/skin colour was higher among individuals with food security (k=0.81; 95% CI 0.78 to 0.83) and lowest among those with severe food insecurity (k=0.69; 95% CI 0.66 to 0.73).

Conclusion The results suggest a tendency towards the whitening of participants from families with food security by interviewers.