Selection bias in cluster randomized trials may threaten the validity of the results. This bias may occur either at the level of the cluster or of the individual. We describe measures for maintaining comparability of intervention groups in a cluster randomized trial of a health education package to reduce dietary salt. The setting was 12 villages of the Ashanti region of Ghana. In total, 1896 villagers between 40 and 75 years of age were selected to take part in the trial using stratified random sampling, based on age and sex. Following individuals' consent and baseline measurements in a pair of villages, villages were randomized to intervention or control arms, stratified for locality (semi-urban or rural). Primary outcomes of the trial were reduction in 24-hour urinary sodium and blood pressure. Of the villagers, 1013 individuals agreed to take part, with a response rate of 53%. The groups were comparable with respect to mean (SD) systolic and diastolic blood pressure (125/74 (27/14) mmHg versus 126/75 (25/14) mmHg) and other outcomes. In conclusion, in this study blind recruitment, aided by randomization in small blocks, and stratified random sampling of the subjects within the clusters helped to ensure comparability of intervention groups, which is vital for the validity of the trial results.