In the analysis of gene expression time series, emphasis has been given on the capture of shape (dis)similarity. A number of proximity functions have been proposed for this task. However, none of them will suitably measure shape (dis)similarity with data containing multiple gene expression time series, unless special data handling is made. In this paper, a symbolical description of multiple gene expression time series, where each variable take as a value a time series, in conjunction with a version of a proximity measure are proposed. In this symbolic approach, the shape similarity of each time series is calculated independently, and aggregated at the end. Gene expression data from five distinct time series are presented to a symbolic dynamical clustering method and a Self-Organising Map algorithm. The quality of the results obtained is evaluated using gene annotation allowing a verification of this proposal?s adequacy.