The immune genes of the major histocompatibility complex (MHC) are characterized by extraordinarily high levels of nucleotide and haplotype diversity. This variation is maintained by pathogen-mediated balancing selection that is operating on the peptide-binding region (PBR). Several recent studies have found, however, that some populations possess large clusters of alleles that are translated into virtually identical proteins. Here, we address the question of how this nucleotide polymorphism is maintained with little or no functional variation for selection to operate on. We investigate circa 750-850 bp of MHC class II DAB genes in four wild populations of the guppy Poecilia reticulata. By sequencing an extended region, we uncovered 40.9% more sequences (alleles), which would have been missed if we had amplified the exon 2 alone. We found evidence of several gene conversion events that may have homogenized sequence variation. This reduces the visible copy number variation (CNV) and can result in a systematic underestimation of the CNV in studies of the MHC and perhaps other multigene families. We then focus on a single cluster, which comprises 27 (of a total of 66) sequences. These sequences are virtually identical and show no signal of selection. We use microsatellites to reconstruct the populations' demography and employ simulations to examine whether so many similar nucleotide sequences can be maintained in the populations. Simulations show that this variation does not behave neutrally. We propose that selection operates outside the PBR, for example, on linked immune genes or on the "sheltered load" that is thought to be associated to the MHC. Future studies on the MHC would benefit from extending the amplicon size to include polymorphisms outside the exon with the PBR. This may capture otherwise cryptic haplotype variation and CNV, and it may help detect other regions in the MHC that are under selection.