Quantitative thermal desorption spectrometry can be carried out if the doses are below limits to avoid (7) the creation of surface disorder, (2) the interaction of gas atoms during the desorption cycle, (3) the retrapping of helium during the desorption cycle, and (4) the multiple occupation of traps by helium. These limits are quantified, and it is shown that in general small amounts of gas have to be detected. The accuracy with which the energy E and frequency factor v can be extracted from a measured 1st order desorption peak is ultimately limited by shot noise in the mass spectrometer. This noise is quantified for both the static and dynamic modes of operation. For the dynamic case, the noise is minimized if Δt/τ≈√2 with Δt the sample time and τ the vacuum time constant.In both modes, the error in the derived value of E has the form ΔE/E = C/No⋅ L/S1/2 ⋅ (1 + β#x0002F;2βc)1/2 with Vo the total number of particles in the peak, L the background leak rate, S the (new concept of) system sensitivity and βc dependent on background pressure, peak temperature, L and No. For the dynamic mode, C is much smaller (10 to 67 times) than for the static mode. The error in E thus decreases monotonically with β and reaches a plateau value for β ≪ βc. Some examples are given showing that accuracies of ∼1% can be obtained for peaks with No ∼ 108 particles for practically achievable values of L and S.