AbstractTo evaluate the accuracy of the effective channel‐length extraction method, the behavior of the measurement errors in the least‐square method, used twice in the extraction procedure, is studied. It is found that the increased number of data are mainly effective in suppressing the dispersion of the measurement data; these data are also somewhat effective in reducing the actual measurement errors. Based on this result, the actual effective channel‐length extraction using experimental MOSFET data is studied. It is observed in the samples used here that the linear relation between total MOSFET resistance and design channel length changes in short channel region. It is found in such a case that as many reference MOSFETs as possible, including both short and long MOSFETs, should be