Based on the related partial least-squares (PLS) algorithms, a new improved recursive partial least-squares (SR-PLS) algorithm was derived. The SR-PLS simplifies the recursive computations by substituting the two principal matrices of the independent variables for the residual matrices directly. By using a computer simulation, it shows the consumed time using SR-PLS decreased greatly. The SR-PLS was implemented in modeling peak resultant cutting force of an NC machining process. The estimated outputs of the process model with the identified parameters were re-constructed, consistent with the measurement. The simulation and experiment demonstrate that the SR-PLS algorithm was correct and effective, with the advantages of less computation time, valid parameter estimation and small modeling errors. SR-PLS algorithm is suitable for the on-line modeling applications in the cases of large-scale measurement data and the requirement of high efficiency as well.