The peripherin gene has three potential ATG translation initiation sites at positions 38, 56, and 290. The second ATG has been proposed to be the initiation codon used for translation of the protein, but there is no experimental evidence for this conjecture. We have isolated a full-length peripherin cDNA (designated as p61-11) from a rat brain cDNA library. Upon sequencing, we found that this cDNA contains a point mutation at the second potential translation initiation codon, which changes this ATG to ACG. When expressed in SW13 cl.2 vim- cells, a cell line without any detectable cytoplasmic intermediate filaments, the protein product of p61-11 cannot form a filamentous network and the major product is 45 kDa in size, which is most likely initiated from the third ATG. The protein product from the first ATG (57 kDa in size) of p61-11 is also detected albeit in smaller amounts. We introduced a frame-shift mutation upstream of the third ATG in p61-11 to create p61-11FS and showed that the third ATG is able to initiate translation efficiently even in the presence of the first ATG, and the 45 kDa protein leads to a diffuse nonfilamentous staining pattern in vim- cells confirming that the first ATG may not be the preferred translation initiation codon, since it cannot suppress a downstream ATG. We increased the translation efficiency from the first ATG of p61-11 by mutating the three nucleotides preceding this first ATG and thereby placing it in a better Kozak consensus sequence for translation initiation. The resulting 57 kDa protein is able to form a filamentous network in vim- cells. We corrected the mutation in the original p61-11 by polymerase chain reaction and generated two peripherin constructs: perM1M2 (which contains all three translation initiation codons) and per delta 1M2 (the first ATG is deleted, but the other two are present). When transfected, their protein products, about 57 kDa in size, form filamentous networks in the absence of other cytoplasmic intermediate filaments. Since there is no 45 kDa protein detected for these latter two constructs, it is reasonable to conclude that in the presence of the second ATG, little or no translation is initiated from the third ATG. Taken together, these results strongly suggest that the second ATG is the preferred translation initiation codon for the peripherin gene.