Relative validity and reproducibility of a semi-quantitative food-frequency questionnaire (FFQ) for measuring preschool children's usual Ca intake were assessed using parents or guardians as a proxy. Estimated diet records (EDR; 3 d) were used as the reference method and reproducibility was measured by repeated FFQ administrations 5 weeks apart. From 2095 preschool children (2.5-6.5 years) randomly selected in Flanders (Flemish region of Belgium), 1052 returned a FFQ and EDR. Stringent exclusion procedures reduced the sample for validity analyses to 509 children. From a separate sample of 244 preschool children, 124 returned two FFQ, of whom sixty were included in the reproducibility analysis. Mean Ca intakes were 838 (sd 305) and 777 (sd 296) mg/d for EDR and FFQ respectively, indicating a mean difference of 60.9 (sd 294.4) mg/d (P<0.001). Pearson's correlation was 0.52. Cross-classification analysis of the FFQ and EDR classified 83 % of the subjects in the same or adjacent category and 2.4 % in extreme quartiles. Actual values for surrogate FFQ quartiles showed a progressive increase in Ca intake (P<0.001). The FFQ correctly identified 77 % of the children consuming less Ca than the age-specific RDA (800 mg/d). Correlation between repeated administrations was 0.79. No significant difference between mean Ca intakes was established by two administrations (23.8 (sd 161.2) mg/d). Cross-classification of repeated administrations classified 93.4 % of the subjects in the same or adjacent category and no subjects in extreme categories. This FFQ tended to underestimate preschool children's Ca intake when administered by a proxy. However, it demonstrated good repeatability and fairly good ability to classify subjects into extremes of Ca intake.