A longitudinal study was conducted to validate a 68-item semi-quantitative food-frequency questionnaire (FFQ) for assessing nutrient intake among pregnant women in rural China. The interview administrated FFQs were conducted at the start of the third trimester (FFQ1) and repeated three months later prior to delivery (FFQ2), and compared with six repeated 24-hour recalls collected between the FFQs. From June to September 2004, 125 women at 23 to 26 weeks of gestation were recruited into the study in two rural counties in Shaanxi province, China. Pearson correlation coefficients for crude nutrients assessed by FFQ1 and FFQ2 had a mean of 0.46, and ranged from 0.31 for thiamin to 0.61 for fat. Pearson correlation coefficients estimated by FFQ2 and 24-hour recalls had a mean of 0.62 for all nutrients, and ranged from 0.53 for cholesterol and carotene to 0.70 for alpha-vitamin E and potassium. De-attenuated coefficients had a mean of 0.69 for all nutrients, and ranged from 0.58 for cholesterol to 0.77 for alpha- vitamin E. Bland-Altman plots revealed significant relationships between the difference and the average of the two methods with larger differences at higher average intakes. On average, 74% of participants were classified into the same or adjacent nutrient intake quintiles, while 1% were grossly misclassified to opposite quintiles. In conclusion, this FFQ has adequate reproducibility and validity for most nutrients but would benefit from the addition of wheat porridge to ensure it is suitable for assessing dietary intake in pregnant women in rural China.