Self-splicing group II intron RNAs catalyze a two-step process in which the intron is excised as a lariat by two successive phosphodiester exchange reactions. The reversibility of the first step has been hypothesized to act as a proofreading mechanism for improper 5'-splice site selection. However, without synthetic access to mis-spliced RNAs, this hypothesis could not be tested. Here, we used a deoxyribozyme to synthesize several branched RNAs that are derived from the ai5gamma group II intron and mis-spliced at the 5'-splice site. Unlike the correctly spliced ai5gamma RNAs, the mis-spliced RNAs are observed not to undergo the reverse of the first step. This is well-controlled negative evidence against the hypothesis that first-step reversibility is a proofreading mechanism for 5'-splice site selection. In a reaction equivalent either to the hydrolytic first step of splicing or to the hydrolytic reverse of the second step of splicing, a mis-spliced 5'-exon can be "trimmed" to its proper length by the corresponding mis-spliced intron, and in one case, the trimmed 5'-exon was observed to proceed correctly through the second step of splicing. These findings are the first direct evidence that this second proofreading mechanism can occur with a group II intron RNA that is mis-spliced at the 5'-splice site. On the basis of the likely structural and evolutionary relationship between group II introns and the spliceosome, we suggest that this second proofreading mechanism may be operative in the spliceosome.