The endotoxin concentration in air was measured in farms where 11 farmers had experienced febrile reactions or allergic alveolitis and in a random sample of farms with 17 symptomless farmers. Samples were obtained during normal dairy farming in eight reference farms (background samples) and in all farms during the handling of material which had probably caused symptoms or disease or, in reference farms, maximal spore exposure (worst case samples). In addition, parallel samplers were used in reference farms, one with a cyclone (5 microns cutoff) and one without, to measure the dust and endotoxin concentrations in the respirable fraction and total dust. The endotoxin worst case values varied from less than 0.01 to greater than 50 micrograms/m3 in symptom farms (median 6.4 micrograms/m3, geometric mean 2.2 micrograms/m3) and from less than 0.01 to greater than 50 micrograms/m3 in reference farms (median 42 micrograms/m3, geometric mean 29 micrograms/m3). This difference was not statistically significant. The background values in reference farms were 1.3 (median) and 0.4 (geometric mean) micrograms/m3. The differences between samples with and without cyclone and between background and worst case samples were statistically significant (p less than 0.02). About 75% of the activity was found in the non-respirable fraction. No correlation was found between exposure to endotoxin and symptoms in farmers. There were weak, but statistically significant, correlations between endotoxin concentrations and total spore count or dust concentrations. The surprisingly high endotoxin values in the respirable fraction of air from environments which apparently did not cause symptoms raises the concern that the Limulus amebocyte assay might be sensitive to other components in the dust rather than endotoxin.