Kelvin probe force microscopy is a scanning probe technique capable of mapping the local surface potential or work function on various surfaces with high spatial resolution. This technique can be realized on the basis of either an amplitude-sensitive method or a frequency-modulation method, which are sensitive to the electrostatic force and its gradient, respectively. We present a detailed experimental and theoretical study of the accuracy and resolution provided by the two methods, including the setup for the frequency-modulation technique. Au(111) with a submonolayer coverage of KCl serves as a test sample exhibiting extended sharply bounded areas that differ in work function by an amount well known from ultraviolet photoelectron spectroscopy. The influence of all relevant experimental parameters on the measurement is investigated. The experimental results are compared with the predictions of a numerical simulation based on a realistic model for the tip-sample geometry. Good agreement is found. The experimental analysis allows us to specify the lateral, vertical, and potential resolution that can be achieved with the two methods for a given tip size. Our work clearly proves that the frequency-modulation method is preferable in most applications because it (i) provides much higher lateral resolution, (ii) yields quantitative surface potential values on areas larger than the tip radius, and (iii) is little affected by variations of the tip-sample distance during topographic imaging.