The combustion gas analysis technique for determination of soil organic carbon (C) incurs substantial running costs and fails to provide information about the type of organic matter. The infrared multivariate calibration approach allows determination of the quantity as well as the quality of C at low cost per specimen but the results will depend on the spectral range chosen and sample collection. The present study examines the calibration models developed on two sets of dry screened soils: one from 20 contrasting agronomic sites in Pennsylvania (n=178), and other from intensive sampling of a single site utilized for a long-term crop-rotation by fertility trial (n=192). The calibration models were developed using partial least squares regression (PLSR) on data collected by two infrared techniques: diffuse reflectance near infrared (DR-NIR) and diffuse reflectance Fourier transform middle-infrared (DR-FTIR). The objectives of the study were to characterize the attributes of different infrared techniques and to identify applications in crop-soil management studies that take advantage of the spectral analytical approach. Our study for the first sets of soils found that models developed with DR-FTIR data performed better with r² of 0.88 than NIR with r² of 0.85 over a C range of 0.28-6.2%, but C prediction errors were higher (0.49 and 0.54 for DR-FTIR and NIR respectively) than the reference method (>0.1%). Prediction errors and biases were also found to be variable with the sampling depths and sites, probably because of the variation in mineral composition. Several organic absorptions are detectable in dry-soil DR-FTIR spectra but not in the NIR, which suggests that the NIR spectral range can be used for calibration of C content, but the approach is unlikely to be useful for molecular characterization of organic matter. PLSR analysis of the second set of soils scanned by NIR and DR-FTIR yielded a much lower C prediction error of 0.17 and 015 with good correlation 0.81 and 0.84 respectively which might be adequate to quantify organic C range found at this site (1.77-3.71). In this case also the mid-infrared range proved to be better for C-calibration models than the near-infrared range. The smaller error value in case of single-site experiment confirms our hypothesis that the calibration models can perform well if built and applied on samples from sites and horizons similar in mineralogy. Our poster will also describe some approaches that can be used to isolate the organic signature to make the calibration models easy to develop using smaller number of variables. We will also try to project an efficient strategy for deciding about the optimum number of calibration samples which should be enough to estimate the organic carbon of the validation set.