The paper illustrates some improvements in the seismic risk assessments in Italy and describes the differences deriving from the use of different approaches to calculate the losses and the influence exerted by different hazard results. The first method of risk evaluation, here termed as 'direct', evaluates the losses by using only the mean values of the main variables involved (rate of events and frequency of the damage levels), thus providing an approximation of the expected losses. The second method, named 'probabilistic', takes into account the uncertainties related to the number of events (hazard) and the damage levels (vulnerability), thus determine the probability associated to each level of loss. Both methods express the risk as the economic losses to dwellings within a reference period of time. Two alternative hazard results are also used to show the influence on the calculated risk: the first one considering the seismicity uniformly distributed within seismic source zones; the second one clustering the strong seismicity in geographically narrowed source zones and scattering the low seismicity over large source zones.The results obtained show that the losses estimated by the 'direct' method are, at national level, a little bit lower than those obtained with the probabilistic method (about 6%). The differences are more pronounced at local level (regional or sub-regional), generally within ±20% with larger values in the zones of lower risk. Nevertheless, also the two hazard results show more pronounced differences at local rather than at national scale. The risk estimates in the high seismicity areas are greater if using the seismic hazard results based on the clustered seismicity, but the reverse is true, in the low seismicity areas, if using the hazard results based on the uniformly distributed seismicity.As a concluding remark, the direct method for calculating losses and the implementation of any seismic hazard result, may be acceptable for a general picture of the risk; whereas, when a detailed description of the territorial distribution of risk is needed, the probabilistic method for computing losses and a well-focused seismic hazard method should be used, as they are more pertinent to describe and highlight local differences.Abbreviations: GdL96 -hazard and risk results from the 'Working Group, 1996'; SCS -hazard and risk results from the 'Spatially Clustered Seismicity' model; SSN -seismic and risk results from the 'National Seismic Survey'.