Methods of calculation of the reliability of process computers are getting more and more important. The paper presents long period observations of 21 process computers with on line applications in the iron and steel industry. About 5000 failure events during 10 years have been systematically recorded and evaluated. On the basis of these long time observations it is possible to develop a probability model of the failure behaviour of process computers._x000D_
This failure model is constructed with four different states of the process computer: Working phase, working phase with undetected reasons, repair phase, failure phase resulting from external influences._x000D_
The model is defined by differential equations for the probability of one of these four phases._x000D_
Long time observations show, that all of these four phases are closely correlated with the capa- city of a process computer, restarting without detection of failure reason, changes of programs and software, costs and time of repair. So the result of this analysis is a model of failure for process computers, which allows to predict breakdowns of process computers on the basis of external parameters.