To develop a dish composition database (DCD) and assess its ability to estimate dietary intake.

The DCD was developed based on 16 d dietary records (DR). We aggregated all reported dishes into 128 dish codes and calculated mean food group and nutrient contents for each code. These data were used to calculate dietary intake in a different population that completed a 4 d DR. The estimated values were compared with those estimated using the standard food composition database (FCD) of Japan.

Japan.

A total 252 adults aged 31–81 years for the 16 d DR (3941 d in total) and 392 adults aged 20–69 years for the 4 d DR (1568 d in total) participated.

There were significant differences in median intakes between the DCD and the FCD for eighteen and twenty (of twenty-six) food groups and for twenty-nine and twenty-two (of forty-three) nutrients (including energy) in men and women, respectively. For food group intakes, Spearman correlation coefficients between the DCD and FCD ranged from 0·19 (animal fats) to 0·90 (fruits and alcoholic beverages) in men (median: 0·61) and from 0·25 (oils) to 0·89 (noodles) in women (median: 0·58). For nutrient intakes, the corresponding values ranged from 0·25 (retinol) to 0·90 (alcohol) in men (median: 0·60) and from 0·15 (retinol) to 0·74 (alcohol) in women (median: 0·53).

Whereas it is difficult to accurately estimate absolute dietary intake values using the present DCD, it has acceptable ability to rank the intakes of many food groups and nutrients.

To compare the relative validity of food group intakes derived from a comprehensive self-administered diet history questionnaire (DHQ) and a brief-type DHQ (BDHQ) developed for the assessment of Japanese diets during the previous month using semi-weighed dietary records (DR) as a reference method.

Between November 2002 and September 2003, a 4 d DR (covering four non-consecutive days), a DHQ (150-item semi-quantitative questionnaire) and a BDHQ (fifty-eight-item fixed-portion-type questionnaire) were completed four times (once per season) at 3-month intervals.

Three areas in Japan: Osaka, Nagano and Tottori.

Ninety-two Japanese women aged 31–69 years and ninety-two Japanese men aged 32–76 years.

Median food group intakes were estimated well for approximately half of the food groups. No statistically significant differences were noted between a 16 d DR and the first DHQ (DHQ1) or between the DR and the first BDHQ (BDHQ1) in fifteen (44 %) and fifteen (52 %) food items for women and in fourteen (41 %) and sixteen (55 %) food items for men, respectively, indicating that both questionnaires estimated median values reasonably well. Median Spearman's correlation coefficients with the DR were 0·43 (range: −0·09 to 0·77) for DHQ1 and 0·44 (range: 0·14 to 0·82) for BDHQ1 in women, with respective values of 0·44 (range: 0·08 to 0·87) and 0·48 (range: 0·22 to 0·83) in men, indicating reasonable ranking ability. Similar results were observed for mean values of the four DHQ and BDHQ.

In terms of food intake estimates, both the DHQ and the BDHQ showed reasonable validity.