Microstructure and welding residual stresses in ferritic heat-resistant steels such as P92 have been considered as one of the most important factors in the structural integrity and life assessment of power plant weldments. Applying computational tools to predict microstructure and residual stress distribution in practical welded structures is a preferable way to create safer, more reliable and lower cost structures. In this work, the effects of volume change, yield strength variation and transformation induced plasticity (TRIP) on the generation of residual stresses in P92 steel welded joints were investigated experimentally and numerically. Optical microscope and Vickers hardness tester were used to characterize the microstructure and hardness of the weldments. The hole-drilling strain-gage method was employed to determine the residual stress distribution across the weldments. Based on SYSWELD software, a thermal-metallurgical-mechanical finite element method (FEM) was developed to simulate welding temperature field and residual stress distribution in P92 steel joints. Firstly, numerical simulations of Satoh test were carried out to clarify the influence of solid-state phase transformation on the formation of residual stresses. The simulation results show that the volume change and the yield stress variation have a great effect on the magnitude and distribution profiles of residual stresses in the fusion zone (FZ) and heat affected zone (HAZ), and even alter the sign of the stresses, while TRIP have a relaxation effect on the tendency of stress variation during phase transformation. Secondly, a FEM was established to calculate the welding residual stress distribution in a single-pass bead-on P92 steel joint. In the FEM, three main constituent phases (austenite, untempered martensite and tempered martensite) in P92 steel were taken into account. Finally, the simulation results of welding residual stress were compared with the experiments obtained by hole-drilling method. The numerical simulation results are generally in a good agreement with the measured data.