This study investigates the dynamic nature of rainfall observed at the Sustainable Agriculture Farming Systems (SAFS) site in California's Sacramento Valley, which was established to study the benefits of winter cover cropping in Mediterranean irrigated-arid systems. Rainfall data of four different temporal scales (i.e. daily, weekly, biweekly, and monthly) are analysed to determine the dynamic nature of precipitation in time. In an arid climate with seasonal precipitation this has large implications for land and water management, both in the short term and in the long term. A nonlinear dynamic technique (correlation dimension method) that uses the phase-space reconstruction and dimension concepts is employed. Bearing in mind the possible effects of the presence of zeros (i.e. no rain) on the outcomes of this analysis, an attempt is also made to compare the dynamic nature of all-year rainfall and winter rainfall. Analysis of 15 years of data suggests that rainfall dynamics at this site are dominated by a large number of variables, regardless of the scales and seasons studied. The dimension results also suggest that: (1) rainfall dynamics at coarser resolutions are more irregular than that at finer resolutions; (2) winter rainfall has a higher variability than all-year rainfall. These results are indeed useful to gain information about the complexity of the rainfall process at this site with respect to (temporal) scales and seasons and, hence, the appropriate model (high-dimensional) type. However, in view of the potential effects of certain rainfall data characteristics (e.g. zeros, measurement errors, scale effects) on the correlation dimension analysis, the discussion also emphasizes the need for further verification, and possibly confirmation, of these results.