FFQ require validation as part of epidemiological research of diet–disease relationships. Studies exploring associations between carbohydrate type and chronic diseases are rapidly increasing, but information on the validity of carbohydrate fractions, dietary glycaemic index (GI) and the glycaemic load (GL) estimated by FFQ is scarce. Likewise, the effects of subject characteristics on FFQ validity have been poorly documented. The present study evaluates the relative validity of an 131-item FFQ in relation to two 3 d food records (FR) performed 6 months apart focusing on the intake of carbohydrate fractions, dietary GI and the GL. Furthermore, we assessed the extent to which subjects' age, education and BMI explain differences between these methods. The study sample comprised 218 men and 292 women aged 25–74 years participating in a large population-based survey in Finland. Energy-adjusted Spearman's rank correlations ranged from 0·27 (sugars) to 0·70 (lactose) for men and from 0·37 (sugars) to 0·69 (lactose) for women. On average, 73 % of the subjects were categorised into the same or adjacent distribution quintile based on the two methods. In general, the FFQ overestimated the intakes compared with FR. Especially in women, FFQ validity for some nutrients was associated with the level of intake, subjects' age and, to a lesser extent, education but not BMI. In conclusion, the FFQ appears to be reasonably valid in the assessment of carbohydrate exposure variables, but the findings show a need for adjustment of diet–disease relationships for subjects' age and education.