【Objective】Iron (Fe) is an essential micronutrient for plant growth. Accurate monitoring of different forms of iron (Total Fe, Available Fe, Free Fe) is essential for soil health management and agricultural production optimization. Total iron and free iron are also necessary indicators to identify certain soil types. Compared with traditional soil determination methods, spectral technology is rapid, cost-effective, and environmentally friendly, and is gradually becoming an alternative for soil Fe determination in recent years. However, there are few reports on the systematic inversion of different forms of iron by spectral technology.【Method】Based on the color parameter (CP), visible near infrared (VNIR), mid infrared (MIR) , and fusion spectrum (SF) were used to obtain data of 501 typical farmland tillage layer (0~20cm) soil samples in Guizhou Province. The spectrum was smoothed by Savitzky-Golay (SG) denoising, and then the baseline was corrected by the standard normalization (SNV) method. Partial least squares regression (PLSR) and support vector machine (SVM) were used for modeling, respectively. The system compared the predictive performance of single spectra and fused spectra for the three types of iron fractions. Moreover, the relationship between accuracy and cost of different prediction strategies was quantified using two indicators: Cost-Efficiency Ratio (CER) and Efficiency Index (EI).【Result】The results show that: (1) in the single spectral model, VNIR spectrum performed best in the prediction of total iron (determination coefficient R2 = 0.85, relative analysis deviation RPD = 2.59, root mean square error RMSE = 5.48 g·kg-1), while MIR spectrum had the highest accuracy in the prediction of free iron (R2 = 0.80, RPD = 2.23, RMSE = 4.15 g·kg-1). Also, the decision level fusion (SF3) spectrum further improved the prediction accuracy of the three forms of iron, among which the effective iron increased the most, but it was still difficult to achieve effective prediction (RPD = 1.37). Thus, using spectral technology to predict soil available iron is not recommended. (2) The cost accuracy analysis showed that the spectral technology can significantly reduce the cost (by 40%~85%) of soil iron characterization. VNIR and MIR technologies had high accuracy and were cost-effective in the prediction of total iron and free iron, and were suitable for scenarios requiring comprehensive consideration of accuracy and cost. However, the accuracy improvement of fused spectrum (SF) was limited, and the cost increased more, which is suitable for scenarios requiring higher accuracy. 【Conclusion】This study shows that spectral technology can significantly reduce the cost of soil iron content measurement on the basis of ensuring a certain prediction accuracy, and can replace the traditional methods to achieve efficient monitoring of total iron and free iron, so as to provide effective technical support for the implementation of precision agriculture.