The utilization of oil shales earns popularity due to the shortage of petroleum of China. However, the clean and efficient conversion of oil shales is limited due to the complexity of the organic matters in oil shales. To understand the structure-reactivity relationship of oil shales in a micro-level, this work studied the distribution characteristics of the covalent bonds of 8 oil shales, and used Artificial Neural Network (ANN) to precisely quantify the relationship between bond distributions, reaction temperature, reaction time and the abundance of free radicals that were generated in pyrolysis and then captured by H from hydrogen-donor solvent. From the sensitivity analysis, the Cal—Cal, Cal—Car, Car—Car, Cal—O, Car—O and Cal=O in oil shales could facilitate the generation and hydrogenation of free radicals in hydropyrolysis. While the Cal· and Car· generated from the breakage of Cal—H and Car—H in oil shales tend to take part in the coupling and condensation reactions. Meanwhile, H· from the breakage of Cal—H and Car-H can capture free radicals, decreasing the hydrogenation effect of hydrogen-donor solvent. And the breakage of O—H of oil shales in the reaction is of low-possibility. The above results to some degrees highlight the mild thermal-conversion technologies of oil shales, coals and biomass.