A large number of patient health data is collected daily in medical information systems. This data contains a non-structural part written in natural language that contains the physician’s notes on specific characteristics of the patient’s medical condition. This section may contain symptoms, diagnoses, therapies, specialties, Latin terms, and other words specific to the medical domain. Useful information suitable for various analyzes could be extracted by processing this section of the text. There are no electronic lexical resources in the Serbian language that are suitable for normalizing and extracting knowledge from medical texts, as well as methods for marking terms in this domain. One reason is that, before any method is applied, the de-identification of patients and staff must be ensured. Also, the evaluation of the results requires manually marked corpora of medical reports in the Serbian language. This paper proposes a method for identifying words belonging to diagnoses in medical texts written in Serbian using natural language processing (NLP) techniques. The proposed method is based on the use of lexical resources, and two set of 1000 medical reports are manually marked for research purposes. In the experimental part, the results of automatic labeling of diagnoses on the marked corpus using the proposed method are presented.