The use of quick attenuated total reflectance infrared (ATR-IR) spectroscopy and near infrared (NIR) spectroscopy to predict extractives content (EC) in heartwood of E. bosistoana with partial least squares regression (PLSR) models was studied. Different spectra pre-processing methods and variable selection were tested for calibration optimisation. While variable selection substantially improved the NIR-PLSR models, only small effects were observed for spectra pre-processing methods and ATR-IR-PLSR models. Both of the NIR-PLSR and ATR-IR-PLSR models yielded reliably EC results with high R2 and low root mean square error (RMSE). NIR based models performed better (RMSE 0.9%) than ATR-IR based models (RMSE 1.6%). Analysis showed that the models were based on IR signals assigned to chemical structures known from eucalyptus heartwood extracts. Combined with PLSR and variable selection, both, ATR-IR and the NIR spectroscopy, can be used to quickly predict EC in E. bosistoana, a measure needed in tree breeding and the quality control of for durable timber.