To have a raw notion of the quality of the SPA solution, RMSEP was evaluated with 100,000 random selections of 5–23 wavelengths a number smaller than 5 is not recommended, as discussed before. Also, if the number is larger than 23—the number of calibration samples—numerical problems arise . in MLR . Remark that an exhaustive search is unfeasible, since the number of possible selections is larger than 8=1016. The result was used to generate an estimated probability density function pdf , which is depicted in Fig. 4. From this graphic, it should be noted that the probability of a random selection being better than the SPA solution is almost zero (area below the pdf curve between 0 and 0.02 mg ). In fact, the best result among the randomly generated selections was RMSEP=0.03 mg .