Plant uptake factor (PUF), single-variable regression of natural log-transformed concentrations in rice grain/vegetables versus natural log-transformed concentrations in soil and multiple-variable regression with soil concentrations and pH, was derived, validated and compared based on the paired crop and soil data collected from studies regarding As, Cd and Pb contaminated croplands in China. Results showed that the median value of PUF did not present deterministic prediction. But after natural logarithm transformation, the PUF followed Gaussian distribution which could be useful in risk assessment. The single-variable regression models were significant for As, Cd and Pb uptake both by rice and vegetables; however, the standard errors of all the regressions were comparatively large. Soil pH as a variable was generally significant but it only contributed positively to model fit for Cd uptake. After model comparison and selection, the upper 95% prediction limits of the multiple regression model for Cd uptake by rice was recommended to calculate screening value of Cd for paddy soil based on the limit for Cd concentration in rice grain.