The objective of this research work was to utilize machine learning tools for predicting the yield and hydrogen contents of bio-oil (H-bio-oil) based on biomass compositions of feedstock and pyrolysis conditions. In this regard, multiple linear regression (MLR) and random forest (RF) method was successfully applied and compared. The results verified RF's larger feasibility than MLR for predicting bio-oil yield and H-bio-oil. Moreover, the profound information behind the model was extracted. The compositions of feedstock exerted more influences on both yield (60%) and H-bio-oil (77%). Besides, the proximate analysis information was preferable to determine yield, which was inverse for H-bio-oil. The modes of each variable affecting yield and H-bio-oil were described by partial dependence analysis. This research provided a reference for upgrading the bio-oil and extended the knowledge into biomass pyrolysis process.