Precursor RNA transcribed from the yeast mitochondrial gene coding for the large ribosomal RNA contains a group I intron that can excise itself in vitro. Apart from group I specific sequence elements the intron also contains a gene encoding a DNA endonuclease involved in intron dispersal. A precursor RNA derivative from which this gene has been removed self-splices efficiently, but due to activation of cryptic opening sites located in the 5' exon, the 3' part of this exon is sometimes co-excised with the intron. Upon further reaction, this enlarged intron molecules give rise to interlocked circles, comprising small circles derived from 5' exon parts and large circles of the intron. Sequence comparison between cryptic opening sites and authentic splice sites reveals in most cases homology with the 3' exon part that is capable of interacting with the Internal Guide Sequence. The role of the IGS was further substantiated by replacing the cryptic opening sites with well defined sequences of authentic splice sites: one corresponding to the 3' splice site and its mutant derivatives, the other to a fragment containing the natural 5'-3' exon junction. Precursor RNAs derived from these constructs give rise to interlocked circles, and mutation studies confirm that the 3' exon nucleotides flanking a 3' splice site are essential for their formation. The results underline the crucial role of the IGS in interlocked circle formation which behaves similarly as in the normal self-splicing reactions. It has been proposed that the two short helices formed by basepairing of the IGS with the 5' and 3' exon can co-axially stack on top of each other forming a quasi continuous RNA double helix or pseudoknot. We present a model explaining how transesterification reactions of a mutant precursor RNA in such a pseudoknot can lead to interlocked circles. The experiments support the notion that a similar structure is also operative in splicing of wild type precursor RNA.