Data sets generated by the Geophysical Fluid Dynamics Laboratory (GFDL) 3-D general circulation/tracer model for an ozone experiment are used to compare the accuracy of various total ozone networks in calculating global and hemispheric means of total ozone and annual trends of monthly mean total ozone. The advantage of such an approach is that model integrals and trends are known exactly, thus providing an ability to examine the errors expected in present and hypothesized sampling networks. The effects of both spatial and temporal sampling errors are presented. Because the 3-D tracer model uses the same time-dependent wind fields from year to year, the influence of interannual meteorological variability and the sampling error resulting from long-term ozone trends cannot be evaluated. Utilizing varying numbers of observations per month, total ozone networks of 9, 53, and 181 stations are compared. In addition, a case of 53+15 new, judiciously placed stations is examined. Model network evaluations of global mean ozone show underestimates of 1--3% occurring because of a compensation of northern and southern hemispheric errors up to -6% and +3%, respectively. An examination of the error of global mean ozone from 9 to 1% for addition of random stations from 5 to 100. Markedly slower improvement is seen with further increases in the number of stations. One-year trend analyses of total ozone are compared for varous networks and individual stations. Sampling errors of nearly 1% per year are seen for the 53 station case, using four perfect, equally spaced observations per month. The errors grow substantially larger with fewer observations. The effect on global and hemispheric means of stations not taking measurments during cloudy periods also is investigated. Results indicate that the weak annual mean cloud bias error (0.285%) is overwhelmed by the larger error produced by the decrease in effective network density.