Stable single-unit recording in the brain of vertebrates allows to investigate processes underlying neural plasticity. In applied aspect long-term single-unit recording can be useful for development of invasive brain--computer interface. Here we propose a criterion for identification of neurons that were recorded for more than one day. Based only on the spike forms classification yields ambiguous result. Additional parameters (such as form of interspike interval histogram or certain parameters of that histogram) decreased misclassification probability considerably. Using proposed criterion we were able to identify 82 neurons that were recoded for more than one day. In extreme case activity of one neuron was observed for 94 days.