This paper presents an application of data mining on aviation incident data in order to predict the level of incidents' seriousness. Every incident can be seen as a problem that must be avoided or at least minimized its consequences. In aviation industry we can identify several interesting tasks that can be solved by means of data mining methods, e.g. prediction of important meteorological phenomena as fog or low clouds; prediction of potential incidents or problem situations etc. In our case we used public dataset from Federal Aviation Administration Accident/Incident Data System containing more than 22 thousand records from the period between years 2000 and 2013. Our goal was to generate a prediction model that will be able to identify possible risk situations based on significant input factors extracted from dataset with the best possible accuracy. This paper describes the whole process as well as the very good results that we achieved. Our model can be further used to reduce the number of incidents with fatal/death consequences.