In this study, we designed and built an automated system of collection and measurement of throughfall and stemflow, developing a new sampling methodology. Throughfall was measured by trough-type system of collectors, each collector with sampling area of 5 cm × 6 m, connected every six troughs to a large tipping bucket raingauge. Our throughfall measurement system covered a larger surface area than do most commonly used randomly relocated gauges, reducing the spatial variability. Temporal resolution was high (5 min), allowing the study of the short-term dynamics of the interception process. Stemflow was collected from 65 trees and also measured by large tipping bucket raingauges. Water vapor exchange at the forest-atmosphere interface was derived from eddy covariance data from a flux tower in the same area as the interception study. During the study period (November 2002-October 2004) a mild El Niño year developed and total annual rainfall was considerably lower than the average for the region. The interception loss in the year with normal rainfall was 13.3%, compared to 22.6% of gross precipitation in the dry year. The interception difference is explained by the comparison of mean intensity and duration of events in the normal year (8.77 mm/h and 1.88 h) versus the driest year (5.36 mm/h and 2.32 h). Interception loss for the whole period represented 16.5% of the gross rainfall, with throughfall 82.9% and stemflow 0.6%. We used the analytical Gash model to estimate the interception loss. The model succeeded in capturing the variability associated to the variability in the characteristics of precipitation. This is the first study to show the variability of interception in relation to rainfall (seasonally and between years). © 2007 Elsevier B.V. All rights reserved.