Interannual‐to‐decadal predictability of the Paraná river in South America is investigated by extracting near‐cyclic components in summer‐season streamflows at Corrientes over the period 1904–1997. It is found that oscillatory components with periods of about 2–5, 8 and 17 years are accompanied by statistically significant changes in monthly streamflow. Autoregressive predictive models are constructed for each component. Cross‐validated categorical hindcasts based on the 8‐yr predicted component are found to yield some skill up to four years in advance for below‐average flows. A prediction based upon the 8‐ and 17‐yr components including data up to 1999 suggests increased probability of below‐average flows until 2006.