Objective:We assessed the validity of a recently developed dish composition database (DCD) against urinary biomarkers compared with a standard food composition database (FCD).Design:Intakes of protein, Na and K were estimated by 2 × 24 h urine collections and by 4 d dietary record data based on the DCD (including 128 dishes) or FCD (including 1878 foods).Setting:Japan.Participants:A total of 161 men and 163 women aged 20–69 years.Results:Compared with the 24 h urine-based estimates, the median intakes estimated using the DCD and FCD differed significantly for protein and Na in men and for Na and K in women. Deattenuated Spearman correlation coefficients using 24 h urine-based estimates for the intakes of protein, Na and K were lower in the DCD (respectively: 0·26, 0·15 and 0·44 in men; 0·22, 0·27 and 0·22 in women) than those in the FCD (respectively: 0·43, 0·40 and 0·59 in men; 0·33, 0·45 and 0·42 in women). When data on dish portion size reported by participants were used for estimation instead of standard portion-size data based on the DCD, the accuracy of the estimated median intakes did not change notably, whereas the deattenuated correlation coefficients improved (for protein, Na and K respectively: 0·32, 0·31 and 0·56 in men; 0·31, 0·41 and 0·39 in women).Conclusions:The DCD supported by individual-level information on dish portion size showed fair to moderate validity in ranking individuals according to their intakes of protein, Na and K, similar to the FCD.