Analysis of symptom lifetime histories (trends) is widely employed in diagnostics of machines, in particular those designed for long service life. These trends, however, often reveal considerable irregularities, or fluctuations, which cannot be attributed to technical condition evolution. They can render diagnostic reasoning vague and uncertain. This results from the fact that symptom values are influenced not only by object condition parameters, but also by a number of other factors, which sometimes are dominant. More detailed analysis leads to a conclusion that a measure of symptom value fluctuations can itself be used as a diagnostic symptom. This conclusion is supported by the model-based consideration, employing a modification of the Energy Processor (EP) model, developed specifically for this purpose. To illustrate this special feature, several examples are presented and discussed, employing databases obtained for large steam turbines.