Ramp calibration has been widely used in ergonomic studies of the shoulder-neck region as a method for estimating muscle forces or glenohumeral torque on the basis of surface electromyographic (EMG) amplitude from the upper trapezius muscle. The relationship between load and EMG amplitude is determined using least-square regression and the present study gives a statistical evaluation of this approach by analysing data from six healthy females performing isometric abductions and flexions of the arm. It is shown that a load estimate corresponding to 25% of maximal strength (25%MVC) may be provided only with a 95% confidence interval of +/- 10%MVC (median of intervals calculated on an individual basis). In many ramps the EMG amplitude residuals of neighbouring points after regression were shown to correlate significantly. This adds further to the length of the confidence interval. 'Short' ramps (load range: 18-31%MVC) generated similar load estimates as 'full' ramps (18-70% MVC), but with a significantly better confidence if inappropriate modelling was controlled for. Ergonomic analyses operating in a real-time domain will be particularly sensitive to errors in pointto-point load estimation and the study suggests that stable results may be obtained only at a group level.