The sizeable fraction of fungal mitochondrial protein-coding genes contain introns of type I or (rarely) of type II, which presents a challenge for their correct prediction. We have developed the annotation pipeline, which for the first time allows the computational prediction of such types of genes. When tested on 82 genomes from GenBank, the algorithm has the accuracy of 91%/88% (sensitivity/specificity) at nucleotide level, and 84%/79% on the exon/ORF level.