The Kalina cycle is believed to be one of the most promising technologies for power generation from low temperature heat sources such as geothermal energy. So far, most Kalina cycle power plants are designed with a working fluid mixture having a fixed composition, and thus normally operate at a fixed condensing temperature. However, the ambient temperature (i.e., heat sink) varies over a large range as the season changes over a year, particularly in continental climates. Recently, a new concept, i.e., composition adjustable Kalina cycle, was proposed to develop power plants that can match their condensing temperature with the changing ambient conditions, aiming at improving the cycle's overall thermal efficiency. However, no detailed analysis of its implementation and the potential benefits under various climate conditions has been reported. For this reason, this paper carried out a comprehensive numerical research on its implementation and performance analysis under several different climate conditions. A mathematical model is firstly established to simulate the working principle of a composition-adjustable Kalina cycle, based on which a numerical program is then developed to analyse the cycle's performance under various climate conditions. The developed numerical model is verified with some published data. The dynamic composition adjustment in response to the changing ambient temperature is simulated to evaluate its effect on the plant's performance over a year. The results show that a composition-adjustable Kalina cycle could achieve higher annual-average thermal efficiency than a conventional one with a fixed mixture composition. However, such an improvement of thermal efficiency strongly depends on the heat source temperature, climate conditions, etc. The composition-adjusting system introduces extra capital and operation costs. The economic viability of a composition-adjustable Kalina cycle power plant depends on the balance between these extra costs and the increase of thermal efficiency. (C) 2016 The Authors. Published by Elsevier Ltd.