It takes quite a long time for an analyzer, such as gas chromatography, to measure a bulk property of a system, which prevents on-line measurements. Also, the cost of installation and maintenance is very high. Consequently, some other means is needed for on-line measurements of properties and the development of soft sensors based on process variables like temperature and pressure is of great interest. In the field of gas industry, the development of a soft sensor which makes indirect on-line measurements of gas compositions and flow rate, is in progress. In this paper, we proposed a hybrid inner model PLS which improved the prediction performance by taking into account the data structure, as an empirical modeling algorithm. When applied to a design of a soft sensor of a distillation tower, the hybrid inner model PLS showed better prediction performance than other methods.