Endplate responses were recorded in frog muscle fibres during an advanced stage of botulinum (BoTX) paralysis, when transmitter release had fallen to a very low level. By simultaneous recording from two points, it was found that, even when the quantal responses had been reduced to less than 0.01 per impulse (that is, four to five orders of magnitude below normal), the release continued to be spatially dispersed along the terminal arborization. These observations make it very unlikely that whole "active zones' could be eliminated, as has been suggested, in all-or-none fashion by local action of BoTX molecules, and they suggest a more graded, indirect mechanism by which the toxin molecules interfere with the sites of transmitter release.