Plant δ13C-temperature (δ-T) relation has been established in many systems and is often used as paleotemperature transfer function. However, it is still unclear about the exact contributions of temperature variation to plant 13C discrimination because of covariation between temperature and precipitation (aridity), which reduces confidence in reconstruction of paleoclimate. In this study, we measured carbon isotope composition (δ13C) of 173 samples of C3 perennial herbs from 22 sites across a temperature gradient along the 400 mm isohyet in the farming-pastoral zone of North China. The results showed that precipitation obviously affected the correlations of temperatures and foliar δ13C. After removing the influence of precipitation by analysis of covariance (ANCOVA), a more strongly positive relationship was obtained between site-mean foliar δ13C and annual mean temperature (AMT), with a regression coefficient of 0.1636‰/°C (p = .0024). For widespread species, Artemisia lavandulaefolia and Artemisia capillaries, the slopes (or coefficients) of foliar δ13C and AMT were significantly steeper (larger) than those of foliar δ13C and AMT where the precipitation influence was not excluded, whereas the δ-T coefficients of Polygonum persicaria and Leymus chinensis showed little change across the transect after deducting the precipitation effect. Moreover, the positive relationship between temperature and δ13C over the transect could be explained by soil moisture availability related to temperature. Our results may afford new opportunities for investigating the nature of past climate variability.