Mesquite pod samples collected in Northeast Brazil were read in natura and pre-processed in NIR Perten® DA 7250 (PerkinElmer, Inc., USA), and pre-processed in NIR FOSS 5000 Nirsystem II using ISIScan® software. Reference analyzes were obtained by conventional methodologies in an animal nutrition laboratory. The reference results showed good variation in chemical composition. In the performance of the models, those developed with sample pre-processing were more efficient than those built with in natura samples. And among the instruments, Perten performed similarly to the FOSS instrument.

Four varieties of forage palm (Gigant, Hand of Maiden or Ipa sertânea, Sweet, Little and Ear of elephant mexican) collected in three states of Northeastern Brazil, were read in natura and pre-dried and ground in a NIR Perten DA 7250 instrument. Models for in natura samples showed satisfactory performance for dry matter and crude protein. Better model deviation performance was given for dry matter for samples in natura, dry matter and crude protein, insoluble fiber in acid detergent and in vitro digestibility of dry matter for samples pre-processed. NIR technology can be used to predict the bromatological chemical composition of forage palm. The pre-processing of the sample proved to be efficient, increasing the robustness of the models for the evaluated nutrients, with the exception of crude protein and dry matter, whose performances were considered as models of excellence, even for the in natura sample.