Soil matric potential quantifies water availability in soils. Low soil matric potentials are difficult to measure with most in situ techniques. This is also the case for the widely-used dielectric MPS-2 sensor. This probe determines matric potential indirectly from the measured water content in its porous sensor ceramics using dielectric permittivity as a proxy for water content. The accuracy of MPS-2 readings was analyzed in desiccation experiments using 13 soils with different texture and organic carbon content and a WP4C dewpoint potentiometer as reference instrument. Further, it was explored whether observed inaccuracies relate to sensor calibration, confounding dielectric effects, or the water release characteristics of the sensor ceramics. Above -1000 kPa, the MPS-2 readings were accurate in all tested soils with a mean deviation of 3% to the reference values. Below -1000 kPa, MPS-2 readings were increasingly higher than the reference in all tested soils, but the deviation from the reference depended on soil type. Poor factory calibration of the sensors, soil texture dependent differences of water flow at the soil-ceramic interface, and dielectric effects are supposed to be the main reasons for the low and soil-type-specific MPS-2 sensor accuracy in dry soils. Nevertheless, the high consistency of the MPS-2 readings allowed us to derive soil-type-specific equations to improve the accuracy of measurements to values as low as -4000 kPa. We recommend applying the equations to any MPS-2 readings below ~ -1000 kPa to obtain more reliable data and thus an improved insight into the role of soil water in ecosystems.