Spending per scientist declined precipitously within African agricultural R&D agencies over the past several decades. In 1991, average cost per researcher across 147 R&D agencies was $119,300 in 1985 international dollars–or US$59,500 when measured in United States rather than international dollars—34 percent below the corresponding 1961 figure. This trend reflects the rapid growth in numbers of scientific staff compared with the slow growth in funds to support them. Comparatively low, and often shrinking, real salaries per scientist are a factor too. African scientists were paid an average of US$5,000 in 1991 (or roughly US$7,500 with fringe benefits included), while comparable average salaries for academic staff working in large public universities in the United States were $58,889 (or $72,667 with fringe benefits included). The new, agency-level data reported in this paper reveal significant variation in the costs per scientist not apparent from the country averages. From the 147 agencies for which we have data, spending per scientist in 1991 ranged from a low of $16,400 for WRRU, Zambia, to $400,000 for ARD, Swaziland (in 1985 international dollars). There were 67 agencies (46 percent) that spent less than $100,000 per scientist per annum. We used some simple econometric procedures applied to a sub-sample of 107 agencies in 21 countries to investigate reason for the large variation in costs per scientist. The intensity of support staff per scientist and the intensity with which expatriate researchers are used are important sources of variation. Larger stations lowered the costs and having more significant influence on its costs. Semipublic agencies typically spent considerably more per scientist than government agencies with 1991 figures of $207,700 for the former, compared with around $104,600 for the latter (in 1985 international dollars). GDP per capita and various other unspecified, country-specific effects also accounted for much of the observed variation in costs per scientist.