Introduction: A microservice architecture, which allows applications to be built as a set of independent microservices that work together to fulfill a common client request, has recently become the basis or even the standard for deploying complex systems that affect multiple physical structures and devices. In addition, the introduction of parallel service scenarios into such systems, especially in the case of high-load systems, makes it possible to increase their efficiency and performance. Purpose: To develop a mathematical model of distributed transactional applications with a microservice architecture and parallel nodes and to evaluate such a performance indicator as the average response time. Results: We present a mathematical model of distributed transactional applications with a microservice architecture in the form of a queueing network with sequential nodes, one of which has a parallel structure with several subnodes, the number of those is more than two. Based on the decomposition method for analyzing queueing networks, an approach to estimating the average response time of the system under consideration is proposed. We use known results for assessing individual nodes of a G/G/1 type network, as well as nodes with division and with parallel service. The results of the computational experiments allow drawing conclusions about the admissibility of the proposed approach, and obtaining recommendations regarding the applicability of various formulas for different load levels, in particular, for those whose average approximation error does not exceed 10%. Practical relevance: The proposed model and the method of its research can be used for the initial assessment and prediction of the average response time of transactional applications with parallel nodes at different load levels and, as a result, can contribute to maintaining the required quality of service for users of transactional applications.