This paper proposes a methodology to study daily precipitation series that include a significant proportion of missing data, without resorting to completion methods based on randomly generated numbers. It is applied to a data set consisting of 75 station records (1951–2000) covering the Italian territory. They are clustered by principal component analysis into six regions: the north-west, the northern part of the north-east, the southern part of the north-east, the centre, the south and the islands (i.e. Sicily and Sardinia). Complete annual and seasonal regional average series are obtained from the incomplete station records, and analysed for droughts and extreme precipitation events. Droughts are identified by means of two indicators: the longest dry period and the proportion of dry days. The most remarkable result is a systematic increase in winter droughts over all of Italy, especially in the north, due mainly to the very dry 1987–93 period. Extreme events are analysed considering 5 day regional totals. In this case, however, an attempt to search for a statistically significant trend is not successful because of the scarcity of events in such a short period. The reliability of the regional series is checked by computing some basic statistics concerning total precipitation, rainy days and precipitation intensity and comparing them with the same statistics computed for regional series obtained by station records completed with methods based on random number generators. Copyright © 2002 Royal Meteorological Society.