An investigation is performed of individual behavioral characteristics of text input that are based on the keystroke dynamics using a virtual keyboard, and their application in the authentication of mobile device users. Key hold time, time between two consecutive keystrokes, and keystroke pressure are used as features to build a user model. A feature vector of a freely typed text is constructed by a user-adapting algorithm that divides digraphs into groups according to the pressing speed and the grouping of keys according to their functionality. Outliers are sought and corrected on the basis of quartiles with subsequent standardization of values. A fuzzy search for exceptions based on potential functions is used to solve the problem of classification when there are no examples of keystrokes by an attacker in a training sample. The resulting software, which consists of a custom Android system keyboard, is used for data collection and experimental assessment of the quality of the proposed approach to building a user behavioral model, proving its effectiveness. The software is also used to develop a prototype of a user authentication system that employs biometric characteristics of interaction with an onscreen keyboard to protect all applications on a mobile device that uses a standard system keyboard.