The visible-near infrared (Vis-NIR) based real-time soil sensor (RTSS) is found to be a great tool for determining distribution of various soil properties for precision agriculture purposes. However, the developed calibration models applied on the collected spectra for prediction of soil properties were site-specific (local). This is found to be less practical since the RTSS needs to be calibrated separately for every field. General calibration approach is expected to minimize this limitation. This paper describes the feasibility of general calibration model developed from two types of paddy field and to compare the performance of the calibration models. For this purpose, Vis-NIR reflectance spectra of fresh soil were acquired at two fields (organic and inorganic paddy fields). Fresh soil samples were also collected from these two fields for analysis of moisture content (MC), organic matter (OM), total carbon (TC) and total nitrogen (TN) in the laboratory.  Three calibration models were then developed for each soil properties using partial least square regression (PLSR) technique coupled with full cross-validation. The first model (CM1) was developed using dataset from organic field, second model (CM2) was from inorganic field and the third model (general model – CM3) was developed from combination of dataset from both fields. The performance of the three calibration models were compared based on the determination of coefficient (R_val²), root mean square error of validation (RMSE_val) and residual prediction deviation (RPD). Results showed for MC and OM, CM3 produced highest prediction accuracy with R_val² of 0.90 and 0.95. For TC and TN, CM1 produced the highest accuracy. CM2 produced the lowest accuracy for all the soil properties. This result could be used as a step towards establishment a robust general calibration model for agriculture soil.