The genetic contribution to autism is often attributed to the combined effects of many loci (ten or more). This conclusion is based in part on the much lower concordance for dizygotic (DZ) than for monozygotic (MZ) twins, and is consistent with the failure to find strong evidence for linkage in genome-wide studies. We propose that the twin data are compatible with oligogenic inheritance combined with a major, genetic or epigenetic, de novo component to the etiology. Based on evidence that maternal but not paternal duplications of chromosome 15q cause autism, we attempted to test the hypothesis that autism involves oligogenic inheritance (two or more loci) and that the Angelman gene (UBE3A), which encodes the E6-AP ubiquitin ligase, is one of the contributing genes. A search for epigenetic abnormalities led to the discovery of a tissue-specific differentially methylated region (DMR) downstream of the UBE3A coding exons, but the region was not abnormal in autism lymphoblasts or brain samples. Based on evidence for allele sharing in 15q among sib-pairs, abnormal DNA methylation at the 5′-CpG island of UBE3A in one of 17 autism brains, and decreased E6-AP protein in some autism brains, we propose a mixed epigenetic and genetic model for autism with both de novo and inherited contributions. The role of UBE3A may be quantitatively modest, but interacting proteins such as those ubiquitinated by UBE3A may be candidates for a larger role in an oligogenic model. A mixed epigenetic and genetic and mixed de novo and inherited (MEGDI) model could be relevant to other “complex disease traits”. © 2004 Wiley-Liss, Inc.