Accurate measurements of soil heat flux are important for energy balance studies on bare soils. Measurements are usually made with passive transducers that transform the vertical soil heat flux into an e.m.f.. Measurement errors or bias result from differences between the calibration coefficients provided by the manufacturer and those determined in the field. These differences result from bad thermal contact between the soil and the fluxmeter and/or changes in soil thermal conductivity. New printed circuit heat fluxmeters are very thin (0.2 mm) and they provide a better thermal contact with the soil because they have an external copper layer instead of an insulating resin. We carried out a theoretical analysis to identify properties of the transducers (geometrical, thermal or electrical) most important for reducing the calibration variability. The transducer thickness was found to reduce the calibration variability due to large soil thermal conductivity variations. Transducer thermal conductivity is also important when the soil thermal conductivity is accurately known. The printed circuit transducers and classical soil heat flux transducers (thermopiles) where then compared in three different soils, a sandy loam, a loamy and a chalky soil under changing climatic conditions in spring. The outputs of both transducers were compared to reference soil heat flux measurements obtained by the heat storage method. The thermopile transducers were more sensitive (4.1 μV W−1 m2) than the printed circuit transducers (1.6 μV W−1 m2). Both transducers gave similar responses when the soil thermal conductivity varied over a narrow range. The total variation of the calibration coefficients of the printed circuit transducer was smaller for all three soils and for days where the soil thermal conductivity varied widely. We conclude that the printed circuit transducers should be used when field calibration is not possible, or when the calibration is not stable following large soil thermal conductivity variations. The experiment also showed that the theory does not completely describe the interaction between calibration coefficients and soil properties. We have therefore developed a new interpretation of the experimental data that takes into account the thermal contact between the soil and the transducer.