The new CARMENES instrument comprises two high-resolution and high-stability spectrographs that are used to search for habitable planets around M dwarfs in the visible and near-infrared regime via the Doppler technique. Characterising our target sample is important for constraining the physical properties of any planetary systems that are detected. The aim of this paper is to determine the fundamental stellar parameters of the CARMENES M-dwarf target sample from high-resolution spectra observed with CARMENES. We also include several M-dwarf spectra observed with other high-resolution spectrographs, that is CAFE, FEROS, and HRS, for completeness. We used a {chi}^2 method to derive the stellar parameters effective temperature T_eff, surface gravity log g, and metallicity [Fe/H] of the target stars by fitting the most recent version of the PHOENIX-ACES models to high-resolution spectroscopic data. These stellar atmosphere models incorporate a new equation of state to describe spectral features of low-temperature stellar atmospheres. Since T_eff, log g, and [Fe/H] show degeneracies, the surface gravity is determined independently using stellar evolutionary models. We derive the stellar parameters for a total of 300 stars. The fits achieve very good agreement between the PHOENIX models and observed spectra. We estimate that our method provides parameters with uncertainties of {sigma} T_eff = 51 K, {sigma} log g = 0.07, and {sigma} [Fe/H] = 0.16, and show that atmosphere models for low-mass stars have significantly improved in the last years. Our work also provides an independent test of the new PHOENIX-ACES models, and a comparison for other methods using low-resolution spectra. In particular, our effective temperatures agree well with literature values, while metallicities determined with our method exhibit a larger spread when compared to literature results.

M-dwarf stars are the most numerous stars in the Universe; they span a wide range in mass and are in the focus of ongoing and planned exoplanet surveys. To investigate and understand their physical nature, detailed spectral information and accurate stellar models are needed. We use a new synthetic atmosphere model generation and compare model spectra to observations. To test the model accuracy, we compared the models to four benchmark stars with atmospheric parameters for which independent information from interferometric radius measurements is available. We used $\chi^2$ -based methods to determine parameters from high-resolution spectroscopic observations. Our synthetic spectra are based on the new PHOENIX grid that uses the ACES description for the equation of state. This is a model generation expected to be especially suitable for the low-temperature atmospheres. We identified suitable spectral tracers of atmospheric parameters and determined the uncertainties in $T_{\rm eff}$, $\log{g}$, and [Fe/H] resulting from degeneracies between parameters and from shortcomings of the model atmospheres. The inherent uncertainties we find are {\sigma}$T_{\rm eff}$= 35 K, {\sigma}$\log{g}$ = 0.14, and {\sigma}[Fe/H] = 0.11. The new model spectra achieve a reliable match to our observed data; our results for $T_{\rm eff}$ and $\log{g}$ are consistent with literature values to within 1{\sigma}. However, metallicities reported from earlier photometric and spectroscopic calibrations in some cases disagree with our results by more than 3 {\sigma}. A possible explanation are systematic errors in earlier metallicity determinations that were based on insufficient descriptions of the cool atmospheres. At this point, however, we cannot definitely identify the reason for this discrepancy, but our analysis indicates that there is a large uncertainty in the accuracy of M-dwarf parameter estimates.