An Information Retrieval approach to the prediction of values of time series is suggested. Instead of trying to model the data, a pattern is searched for in a history file serving as "text", and relevant occurrences are used as basis for extrapolation. The application is to meteorological data, providing short-term predictions for certain weather features. The experiments show that the algorithm has a good ability of predicting the requested information, and that the predictions are far more accurate than those provided by an analytical prediction procedure, based on approximations using Fourier transforms.