Bad smells are indicators of inappropriate code design and implementation. They suggest a need for refactoring, i.e. restructuring the program towards better readability, understandability and eligibility for changes. Smells are defined only in terms of general, subjective criteria, which makes them difficult for automatic identification. Existing approaches to smell detection base mainly on human intuition, usually supported by code metrics. Unfortunately, these models do not comprise the full spectrum of possible smell symptoms and still are uncertain. In the paper we propose a multi-criteria approach for detecting smells adopted from UTA method. It learns from programmers preferences, and then combines the signals coming from different sensors in the code and computes their utility functions. The final result reflects the intensity of an examined smell, which allows the programmer to make a ranking of most onerous odors.