To address the problems associated with molecular conformations and alignments in the 3D-QSAR studies, we have developed the Flexible Ligand - Atomic Receptor Model (FLARM) 2.0 method. The FLARM 2.0 method has three unique features as compared to other pseudoreceptor model methods: (1) the training ligands are flexibly optimized inside the receptors to achieve minimal docking energies; (2) the receptor atoms are spatially moveable in the process of genetic evolving in order to avoid improper initial receptor shapes; and (3) void receptor sites are specially favored in order to obtain open receptor models that allow large gaps. Advantages of an open model include less noise information, a smaller risk of overfitting, and ease of locating the key interaction sites. The latter two features, inherited from the previous FLARM 1.0 method, can improve the predictive ability of the 3D-QSAR models, while the first feature is newly implemented to relieve the uncertainty caused by improper conformation and alignment. Three FLARM 2.0 case studies were performed, and the results show that FLARM 2.0 models are highly predictive and robust. FLARM 2.0 pseudoreceptor models can correspond well with the pharmacophore models and/or the binding sites of the real protein receptors.