Most Software process models are predefined. When applied in changing environments, they have to be adapted manually. To this end, this paper proposes an adaptive multilateral negotiation model for software processmodeling, namely AMNM-PA. AMNM-PA uses Agents to represent the entities involved in software processes, suchas organizations, teams, persons, etc. and dynamically and adaptively constructs software process models for givensoftware projects by negotiating among the Agents. AMNM-PA is based on non-stationary finite-horizon Markovdecision processes and uses the model-independent Q learning algorithm to choose negotiation strategies, thussupports the dynamic and adaptable negotiation in changing and unknown environments meeting the requirementfor environmental adaptability of the software process modeling. AMNM-PA has been implemented in the softwareprocess management system SoftPM.