© 2019 IEEE.Optimization of the amplitude threshold in extracellular neural recordings has recently become an active research topic in the brain-machine interface literature. In a previous study, the threshold that allows for the encoding of behavioral variables in neural activity with maximum signal-tonoise ratio has been proposed as a reasonable choice. Another good candidate, however, is the maximum likelihood estimate of the threshold. Here, these two types of threshold are estimated using extracellular recordings collected from the motor cortex (M1) of two rats performing a well-learned visuomotor task. The performance of the threshold estimates is assessed by using them in decoders. It is found that, among the four decoders examined, the method that has the best sensitivity, specificity and accuracy is logistic regression that uses the maximum likelihood estimate of the threshold. These results are important for improving the efficiency of brain-machine interfaces.