Infrared spectroscopy has the capacity to predict soil organic carbon (SOC) and total nitrogen (TN) at local/regional scales, but no studies have been conducted to evaluate this technique at a large (cross-regional) scale in Canada. In this paper, mid-infrared (MIR) and near-infrared (NIR) spectroscopies in combination with partial least-squares regression (PLSr) were used to predict SOC and TN in whole soil and in particulate organic matter (POM) fractions on cross-regional, regional, and local scales. Both MIR- and NIR-PLSr models have well estimated SOC [coefficient of determination (R²) = 0.79–0.92, residual prediction deviation/ratio of prediction to deviation (RPD) = 2.19–3.47], TN (R² = 0.70–0.92; RPD = 1.83–3.50), POM-C (R² = 0.76–0.96; RPD = 2.04–5.25), and POM-N (R² = 0.70–0.97; RPD = 1.83–5.78). The prediction efficiency of cross-regional models (R² = 0.90–0.96; RPD = 3.13–5.49) was similar to or better than the prediction of regional (R² = 0.70–0.97; RPD = 1.83–5.78) and local models (R² = 0.70–0.96; RPD = 1.83–5.33) and overall MIR-PLSr models (R² = 0.90–0.96; RPD = 1.98–3.47) yielded similar predictions for SOC and TN relative to NIR-PLSr models (R² = 0.70–0.92; RPD = 1.83–3.50) at cross-regional scale. Hence, it may be possible to develop MIR and (or) NIR spectral models to estimate and monitor SOC, TN, POM-C, and POM-N, and therefore, soil quality, in a rapid and cost-efficient manner across regions with diverse soil types, climate, and cropping history.