Taking the in situ measurements as the driver, the traditional remote sensing models modeling approaches emphasize particularly on the least total errors between the modeled estimate and the measurements while ignore its local error status, which may lead to a large warp between the modeled prediction and the observation at some position. Considering such default of general approach, the present paper developed the subsection mapping retrieval algorithm (SMRA), which decomposes the mapping mechanism between the water qualities and its optical parameters into several subsection functions, and each subsection function is determined by the in situ measurements (named nodes as follows) and an interpolating function. The analysis results of subsection mapping retrieval algorithm based on Newton interposing algorithms indicate that the algorithm keeps the inversion results accuracy at nodes, and is preferably suitable for regression estimate of the complicated relationship between the parameters. Additionally, the method has a great theoretical meaning for the standardization of the sampling interval and sample number in water qualities experiments. Combined with the analysis results of the Landsat/TM imagery and the experimental data of Taihu Lake, it could be known that the SMRA is preferably suitable for describing the relationship between remotely sensed parameters and water qualities, especially for complicated case Ⅱ water bodies such as Taihu Lake.