The average RMSEP obtained was 0.02 mg ly1, which matches the value yielded by SPA. The best set of selected wavelengths indicated in Table 1 resembles the SPA-selected set. Remark, however, that obtaining 100 GA realizations took about 3300 s, while SPA spent only 16 s in the same microcomputer. These times do not include the pre-selection of the spectral window, nor the search for the best number of wave lengths. However, if the computational effort required by the pre-selection procedure were to be taken into account, it would have to be added both to GA and to SPA. It could be argued that GA does not require pre-selection steps, nevertheless the use of full spectrum would increase the dimension of the search space, thus decreasing the probability of GA finding a good solution in a given number of realizations. The same line of reasoning applies if GA is not provided with the number of wavelengths to be selected.