The Compton camera is a gamma ray imaging device expected to provide clinically relevant images in the SPECT applications where collimated cameras are suboptimal. Its imaging performances depend not only on the design of the detection system but also on choices related to tomographic reconstruction. The aim of this work is to show that the accuracy in modeling the acquisition largely influences the quality of the images. For this purpose we restrict here to Doppler broadening models in conjunction with the list-mode maximum likelihood expectation maximization (LM-MLEM) algorithm. The study was carried out with Monte Carlo simulation. We show that the reconstructed point spread function is location-dependent when the model is not accurate, and the usual elongation artefacts well-known in Compton camera imaging will appear. The model we propose allows to reconstruct isolated point sources and more complex non-uniform sources with improved resolution even in the direction orthogonal to the camera.