A total of 50 soil samples were collected from a Danish field varying in texture from sandy to loamy soil. Near-infrared reflectance (NIR) spectroscopy and mid-infrared reflectance (MIR) spectroscopy combined with chemometrics methods, were used to predict the soil texture (clay, sand and silt) and soil organic carbon (SOC). The objective of this study was ：to combine NIR and MIR to predict soil properties in the sampled field. The secondary objective was (i) to use principle component analysis (PCA) and iterations of calculation to find the optimal number of replicates for MIR measurement; (ii) to use partial least square regression (PLSR) in combination with jack-knifing on NIR and MIR spectral data in order to predict soil properties and to find relevant spectral information. The present study showed that using repeated MIR measurements it was possible to reduce the root mean square error of cross validation (RMSECV) by 20-25%. The optimal number of MIR measurement replicates for SOC was 2, while 4 replicate measurements were necessary for soil texture; The lowest predictive errors on a validation set were RMSECV_soc=0.35, RMSECV_clay=0.92, RMSECV_silt=0.81, RMSECV_sand=1.8 using PLSR in combination with jack-knifing. The conclusion is that all models have a good and to a large extent comparable errors. PLSR with jack-knifing simplified and enhanced the interpretation of the developed models due to a reduction in the number of variables used in the developed models.