A magnetic force exhibits a strongly nonlinear property and is usually modelled by a simple function proportional to (pole-gap distance)-P where P > 0. If the function is to be rigorously calculated mathematically however, a great deal of effort may be required. This paper discusses an algorithm and presents program to identify numerically the magnetic force in the form : F(x)=A/(B+x)P+C, in which constants A, B, C and P are determined by processing a set of discrete data points to satisfy the least square error condition. The listing of the sub-program is given together with an example of the main program which drives the conversational computation. An example of the conversation is recorded and numerical example is demonstrated, in which the case of a remarkably nonlinear spring (P=7∼8) is found experimentally in a pair of super-permanent magnets whose material is composed of Nd2Fe14B. The identification error of the formula is less than 3% of that of the conventional formula when B=C=0 formally.