The aims of this study were to assess the relative validity and reproducibility of a semi-quantitative food-frequency questionnaire (FFQ) applied in a large region-wide survey among 2.5-6.5 year-old children for estimating food group intakes. Parents/guardians were used as a proxy. Estimated diet records (3d) were used as reference method and reproducibility was measured by repeated FFQ administrations five weeks apart. In total 650 children were included in the validity analyses and 124 in the reproducibility analyses. Comparing median FFQ1 to FFQ2 intakes, almost all evaluated food groups showed median differences within a range of ± 15%. However, for median vegetables, fruit and cheese intake, FFQ1 was > 20% higher than FFQ2. For most foods a moderate correlation (0.5-0.7) was obtained between FFQ1 and FFQ2. For cheese, sugared drinks and fruit juice intakes correlations were even > 0.7. For median differences between the 3d EDR and the FFQ, six food groups (potatoes & grains; vegetables Fruit; cheese; meat, game, poultry and fish; and sugared drinks) gave a difference > 20%. The largest corrected correlations (>0.6) were found for the intake of potatoes and grains, fruit, milk products, cheese, sugared drinks, and fruit juice, while the lowest correlations (<0.4) for bread and meat products. The proportion of subjects classified within one quartile (in the same/adjacent category) by FFQ and EDR ranged from 67% (for meat products) to 88% (for fruit juice). Extreme misclassification into the opposite quartiles was for all food groups < 10%. The results indicate that our newly developed FFQ gives reproducible estimates of food group intake. Overall, moderate levels of relative validity were observed for estimates of food group intake.