I agree with Jeff Boyd that there is little statistical methodology for
classification patterns. For the most part, this is the result of
the noise-like nature of EMG signals, which makes reproducability a

One way to obtain representative patterns is to calculate an ensemble
average of cyclic movements, such as walking. A useful literature
referencemight be:
Winter D.A. (1984) Pathologic gait diagnosis with computer-averaged
electromyographic patterns. Arch. Phys. Med. Rehab. 65:393-398.

This is basically the method that is used in our laboratory for EMG
research in the horse. There is very little known about the 'normal'
EMG patterns in horses, and we have only studied normal locomotion
until now. There are significant differences between individuals
however, which manifest itself as extra activation periods of certain
muscles. For instance, we see that some horses actively flex the hoof
in the swing phase, while others just let inertia do the work. This
suggests that clinical applications should be possible. We have no
method (yet) to quantify the differences, we only look at the pattern.
Winter's paper indicates the problems associated with automated
parameter extraction from EMG patterns, and also gives some possible

This is our standard method to obtain the EMG patterns:
1. Surface electrodes, 20mm separation.
2. Preamplifier (100x), carried on the subject. Connected by 25m cable
to recording equipment.
3. Amplifier (up to 500x), bandpass filter (30-1000 Hz).
4. Full-wave rectifier, lowpass filter (30 Hz).
5. Analog to digital converter (12 bit), sampling interval 3 ms.
6. Ensemble averaging with interactive graphics display for selection
of strides. Uses signal from hoof-mounted accelerometer for
automatic stride detection.
7. Optional smoothing by a digital filter.

Ton van den Bogert
Dept. of Veterinary Anatomy
University of Utrecht, Netherlands.