Near infrared reflectance spectroscopy (NIRS) is currently being tested in soil analysis but it has rarely been used on contrasting soils to predict soil N supply. We assessed NIRS prediction of soil properties related to N availability [NH₄, NO₃, total N, total C, C/N ratio] and field-based indices of soil N supply [corn (Zea mays L.) N uptake and yield], and how predictions were affected by sample set homogeneity. The soil samples were collected from 2000 to 2009 from 48 sites across four Canadian provinces (British Columbia, Ontario, New Brunswick and Quebec). The total sample set (n = 282) was divided into fine- (≥35% clay; n = 101) and coarse-textured (<35% clay; n = 181) sample sets. Modified partial least squares regression was used to develop prediction equations based on 80% of samples in the calibration set. These equations were validated using the remaining samples and evaluated using the coefficient of determination (R²) and ratio of performance deviation [RPD = SD/SEP(C)]. For the total set of samples, predictions were reliable for total N, total C and C/N ratio (0.7 ≤ R² ≤ 0.9, 1.75 ≤ RPD ≤ 3), and less reliable for NH₄, NO₃, N uptake and yield (R² < 0.7, RPD < 1.75). Prediction accuracy for total N, total C, C/N ratio, NH₄ and N uptake increased with set homogeneity (i.e. from total to texture sets) and were better for the fine-textured set, except for N uptake which was reliable (RPD = 2.46) only for the coarse-textured set. This study demonstrated the possibility of developing reliable NIRS predictive models for soil total N, total C, and C/N ratio from contrasting soils and climatic conditions. Reliable NIRS prediction for NH₄ and N uptake can also be developed when soil samples are grouped based on texture.