The ribosome rapidly translates the information in the nucleic sequence of mRNA into the amino acid sequence of proteins. As with any biological process, translation is not completely accurate; it must compromise the antagonistic demands of increased speed and greater accuracy. Yet, reading-frame errors are especially infrequent, occurring at least 10 times less frequently than other errors. How do ribosomes maintain the reading frame so faithfully? Geneticists have addressed this question by identifying suppressors that increase error frequency. Most familiar are the frameshift suppressor tRNAs, though other suppressors include mutant forms of rRNA, ribosomal proteins, or translation factors. Certain mRNA sequences can also program frameshifting by normal ribosomes. The models of suppression and programmed frameshifting describe apparently quite different mechanisms. Contemporary work has questioned the long-accepted model for frameshift suppression by mutant tRNAs, and a unified explanation has been proposed for both phenomena. The Quadruplet Translocation Model proposes that suppressor tRNAs cause frameshifting by recognizing an expanded mRNA codon. The new data are inconsistent with this model for some tRNAs, implying the model may be invalid for all. A new model for frameshift suppression involves slippage caused by a weak, near-cognate codon.anticodon interaction. This strongly resembles the mechanism of +1 programmed frameshifting. This may mean that infrequent frameshift errors by normal ribosomes may result from two successive errors: misreading by a near-cognate tRNA, which causes a subsequent shift in reading frame. Ribosomes may avoid phenotypically serious frame errors by restricting apparently innocuous errors of sense.