ABSTRACT Isolates of citrus tristeza virus (CTV) differ widely in their biological properties. These properties may depend on the structure of viral RNA populations comprising the different isolates. As a first approach to study the molecular basis of the biological variability, we have compared the sequences of multiple cDNA clones of the two terminal regions of the RNA from different CTV isolates. The polymorphism of the 5' untranslated region (UTR) allowed the classification of the sequences into three groups, with intragroup sequence identity higher than 88% and intergroup sequence identity as low as 44%. The variability of an open reading frame (ORF) 1a segment adjacent to the 5' UTR supports the same grouping. Some CTV isolates contained sequences of more than one group. Most sequences from Spanish isolates belonged to group III, whereas a Japanese isolate was composed mostly of sequences of groups I and II. The mildest isolates contained only sequences of group III, whereas the most severe isolates also contained sequences of groups I, II, or both. The most stable secondary structure predicted for the 5' UTR was composed of two stem-loops and remained essentially unchanged as a result of compensatory mutations in the stems and accommodation of most of the variability in the loops. In contrast to the 5'-terminal region, the variability of the 3'-terminal region of CTV RNA was very much restricted, with nucleotide identity values higher than 90%. The presence of a conserved putative "zinc-finger" domain adjacent to a basic region in p23, the predicted product of ORF 11, suggests that this protein might act as a regulatory factor during virus replication.