Many researchers in Poland and abroad were investigating the problem of machinery diagnosing, which work under condition of varying external load, but none has solved the problem. The researchers were concentrated on laboratory investigations and under condition of simplified form of load varying. The authors showed the solution of the problem and verified it on real objects, which characterized with the complex design and fast varying external load with possibility to reduce to one second variations. In the author’s results of investigations it is stated that there is not one point which is the signal represent but it is a reference function. It is called yielding characteristic, which shows the influence load variation on the characteristic. When preparing the characteristic one has to take into consideration the external load variation or the velocity variation and coming from this using short time signal estimators. To obtain the characteristic one ought to make the regression analysis for signal estimators. From the analysis the linear regression has obtained, for which parameters of the regression as the value of a slope and a vertical coordinate intercept are obtained. At beginning as a machine condition is worsen the slope of regression line increasing. It is caused by bearing condition change by friction wear or a bearing race fault. In both cases it causes shafts run under misalignment. Under the varying external load condition the parameters of the vibration signal is the function of the external load and function of machine condition change.