Spotting by firebrands is the only mechanism that is capable of initiating fires miles from a fire line. The firebrand phenomenon can be understood in three major sequential phases: production, transportation, and ignition of the recipient fuel. Among all, the production is the least studied phase. This research analyzes the generation of firebrands from selected structural fuels both experimentally and theoretically. By establishing generation and characterization of firebrands as the main goal, the flowing objectives were achieved. First, the Bayesian statistics is employed to analyze the available data. After utilizing the power analysis the required sample size, which can represent the characteristics of the entire population, was found to be 1400 for each experiment. Secondly, forty five full scale building components (wall and corner assemblies, and fences) were set on fire in a wind tunnel with differing wind speeds and the generated firebrands were collected. In total, 63 thousands firebrands were accumulated; thus, the conventional measurement techniques were no longer practical. Thirdly, to ease the measurement process, a unique image processing algorithm was developed and the results was used as the input of neural networks and the Gaussian regression process. This created predictive models that estimate the mass of each firebrand within 5% error instead of physically weighing each firebrand. Fourthly, in order to assess the lethality of firebrands, a unique setup was designed to monitor the variation of the surface temperature of smoldering firebrands at different wind speeds with three instruments.