Mid-infrared (MIR) spectroscopy has been established as a rapid and cost-efficient laboratory analysis technique for soil organic carbon (SOC) concentrations. Current portable, high-performance handheld MIR instruments have opened the opportunity to take the technique from the laboratory to the field. We tested the potential of handheld MIR spectroscopy for SOC estimation with field spectral data against parallel VIS-NIR measurements and further evaluated the impact of soil state (in situ, dried, ground) on the accuracy of SOC prediction models. MIR and VIS-NIR data were collected for a set of 90 soils from 90 agricultural loess sites in Central Germany in the field, (i) in situ on the soil surface, and in the laboratory, on (ii) dried and sieved (¡ 2 mm), and (iii) dried, sieved and ground sample material (similar to 10 mu m). Multivariate calibrations for SOC for each pre-treatment were evaluated with a repeated double cross-validation (rdCV) analysis and the Kennard-Stone (KS) calibration sampling approach. MIR calibrations were more accurate than VIS-NIR calibrations for samples measured under the same conditions in each case with lower RMSE values from 27 to 56% in the rdCV and from 15 to 61% in the KS approach. Sample pre-treatment in the laboratory had a pronounced effect on calibrations in the MIR as models developed on finely ground samples (R-rdCV(2) = 0.86, RMSErdCV = 0.11%) were more accurate than those for dried and sieved samples (R-rdCV(2) = 0.79, RMSErdCV = 0.13%), but differences in the VIS-NIR were negligible. SOC estimates with in situ MIR measurements were less accurate due to spectral variation induced by surface heterogeneity and soil moisture but compared favourably against VIS-NIR data (R-rdCV_MIR(2) = 0.63, RMSErdCV_MIR = 0.17% vs R-rdCV_VIS-NIR(2) = 0.39, RMSErdCV_VIS-NIR = 0.23%). Our findings show that portable MIR spectroscopy can achieve superior SOC calibrations to the VIS-NIR range with in situ data demonstrating its applicability for the on-the-go acquisition of spectral soil information in the field. Handheld MIR instruments thus have the potential to facilitate quantitative applications in proximal soil sensing, although with less accuracy than achievable with pre-treated samples in the laboratory.