This session was focused on recent advances of ab-initio structure determination in the macromolecular field, covering different resolution ranges from very high atomic resolution to low resolution. At the high resolution end, the landmark work of the groups of Herbert Hauptman (Shake and Bake, by Weeks, Xu, Miller and Hauptman) and George Sheldrick (Half Baked, by Uson and Sheldrick) was presented by Chuck Weeks and Isabelle Uson, respectively. In this work, the power of reciprocal space direct methods is combined with real space filtering. The net result is the possibility of solving macromolecular structures with as many as 1000 independent non-H atoms, provided that data to 1.1-1.2 A resolution is available. These limits can be pushed further if some heavier atoms are present; for example, Hirustasin has been solved by the group of G. Sheldrick using 1.4 A data by using the 5 disulphide bridges. These programs can also be used for related problems; for example, Shake and Bake has been used for the determination of the Se substructure from anomalous data for structure determination using selenomethionine derivatives. In this case, 3 A data is more than adequate. When very high resolution data is not available, another techniques can be useful. Such is the case of the ARP ( Automated Refinement Procedure), developed by Lamzin, Morris and Perrakis and presented by Victor Lamzin. This technique is useful when some electron density map is already available, and is based in the iterative interpretation and refinement using an atomic model, which is automatically updated. It was shown that this technique can work for cases where the original phase information is limited, and theoretically it could approach the ab-initio case. The method uses the macromolecular stereochemistry to overcome the problem of limited resolution, and can work with 2 A resolution data. A different approach to the limited resolution problem was presented by Chris Gilmore, who in his work with Wei Dong efficiently samples the phase space using error-correcting codes. These codes allow for an appropriate sample with a manageable number of phase sets ( of the order of a few thousands) , which can then be ranked using a maximum likelihood criteria. For the case of crambin a correct solution ( map correlation of 0.54, phase error of 37.5°) was found in the top 8 selections. A similar problem of sampling is present at the very low resolution end, where the approach of Lunin, Lunina, Petrova, Urzhumtsev and Podjarny, presented by Alberto Podjarny, uses modeling in terms of very few atoms to diminish the number of phase sets to be analyzed. In this Few Atoms Model ( FAM) algorithm, a large number of models ( a few millions) is generated, of which a few thousand are selected by maximum likelihood criteria and grouped in a small number of clusters, one of them being close to the right solution. Application of this algorithm to the T50S ribosomal particle, using recently developed approaches for phase extension and cluster selection, resulted in images which have a high ( 0.66) map correlation with corresponding EM images. Last but not least, the new theoretical approaches to the phase problem were represented by the work of Giacovazzo, Siliqi, Fernandez-Castano and Comunale, presented by Carmelo Giacovazzo, who developed the phase relationships for structure factors with rational indices. This method encompasses previous results obtained via the Hilbert transform and the sampling Shannon theorem. This session was very timely, as it came at a moment where important results are emerging. Clearly, the most striking results come at the very high resolution range, where the phase problem is highly over-determined and the atomicity constraint can play at its full strength. However, important advances have been made also at another resolution ranges, where the combination of pseudo-atomic modeling, maximum likelihood ranking and efficient sampling techniques open new possibilities.