Prediction of soluble solids in pulp of 92 samples of guava (Psidium guajava L.) and (Psidium friedrichsthalianum) in two maturity stages was made by using NIRS technique (near infrared reflectance spectroscopy). Each fruit was processed to obtain the edible fraction in which it was determined the soluble solids content (%). Two subsamples were taken and scanned in NIR spectrophotometer at the range of 400 to 2500 nm. Calibration regression models were generated by using MPLS (modified partial least squares) at the range of 1108 – 2498.2 nm. For the choice of the final model of the NIR calibration the criteria considered were the coefficient of determination (R2) and the residual predictive deviation (RPD). Analysis of variance and mean test employing the statistical program SAS v 9.0 were realized in order to establish differences in the results obtained by the primary or reference method and NIR. The results showed a high predictive power of the model (R2 = 0.990) when compared the primary and NIR measurements. The RPD value was 6.20, indicating excellent prediction accuracy. Primary estimates and NIR not showed significant differences. Ho wever there were significant differences at level of samples.