This paper is concerned with the statistical investigation of a massive data bank of 49 years of rainfall rate R continuously recorded in Barcelona using a rain-rate gauge with ten seconds response time. With radio communications in mind, the paper addresses and reviews in detail: 1) reliable statistical model for R, 2) number of years required to obtain a database from which to derive a reliable R-distribution, and 3) the CCIR worst-month concept. The research has shown that the generalized Pareto a exp (- pR)/Rb gives nearly perfect fit for all ranges of R followed closely by the gamma distribution, and the simpler square root (R”*) normal distribution gives excellent fit too. The log-normal distribution was unsatisfactory for R 5 60 mm/h. The spread of the yearly distribution of P(R) is cube root normally ‘distributed ([f(R)]”3) and between 7 and 10 years are required before a reliable average distribution P(R) can he obtained. The study of the P(R) return time in years is also presented. High resolution of P(R) is presented looking at the evolution of the annual P(R) in terms of the hourly and monthly contributing parts revealing statistical features such as the location in time of rain rates above 50 mm/h. Finally, the study shows that the calendar month contribution to P(R) remains at all times well below the synthetic CCIR worst month and recommendations are then given about its use.