A fast response rain gauge, able to record rain intensity down to a step time of 30 seconds and a resolution on rain depth of about 0,02 mm, was developed by the CSTB. With this rain gauge, large series of rain episodes was collected during a period of 2 years and 7 months on the CSTB site of Nantes. The collected data, representing 1345 rain episodes with depth up to 46 mm, duration up to 28 h and 10 minutes mean intensity up to 71 mm/h, were used to define statistical laws representing satisfactorily the average characteristics of the episodes of precipitation; such as distributions of occurrence frequency of the following parameters: global rain quantity, duration and mean intensity of each episode. Weibull law and log-normal law are used to fit these distributions. Characteristics of precipitation intensities on 30 seconds time steps within each rain episodes are also described by mean of log-normal law fitting on cumulative frequency distributions. Values of the descriptive parameters of the log-normal fitting laws describing each episode seem to show some interesting constant features as asymptotic behaviour with rain depth of the episode; information is also given on null intensities occurrence during episodes. A first insight is also given on the autocorrelation values tendencies during episodes, when possible. It seems to confirm the independency of rain episodes after one hour of dry period. This statistical characterization of rain can thus directly be used for numerical and wind tunnel modelling of the rain on the scale level of a building. For example, it can be used to define the hydric behaviour of a facade under an episode of driving rain, the behaviour of a green roof or drain pipe system for water storage. The next research steps will consist, on the one hand to broaden the data base of fine temporal rainfall data on new sites, to complete the statistical average events analysis, on the other hand to approach the study of extreme precipitation events, especially at a very fine time step; but also to start a fractal analysis of the time series.