The technique of surface nuclear magnetic resonance (surface-NMR) provides information on porosity and hydraulic conductivity that is highly valuable in a hydrogeological context. However, the applicability of surface-NMR is often limited due to a bad signal-to-noise ratio. In this paper we provide a detailed insight into the technique of harmonic noise cancellation based on remote references to improve the signal-to-noise ratio. We give numerous synthetic examples to study the influence of various parameters such as optimal filter length for time-domain approaches or the necessary record length for frequency-domain approaches, all of which evaluated for different types of noise conditions. We show that the frequency-domain approach is superior to time-domain approaches. We demonstrate that the parameter settings in the frequency domain and the decision whether or not to use separated noise measurement depend on the actual noise properties, i.e., frequency content or stability with time. We underline our results using two field examples. netic noise. Such techniques have been used e.g., in magnetotellurics (MT) and transient electromagnetics (TEM) since the 1980s (Gamble et al. 1979a,b;Goubau et al. 1984;Spies 1988). A similar approach has become popular with the latest developments in surface-NMR instrumentation that provides the necessary multichannel data acquisition systems (Radic 2006;Walsh 2008).Several methods using a reference loop based system have been developed. Radic (2006) presented a remote reference approach operating in the frequency domain and using vertical loops as references. Only little information is available about the technique implemented in Vista-Clara software (Walsh 2008). Müller-Petke and Yaramanci (2010b) presented a time-domain approach based on a single reference loop and optimal filtering that has been extended to an arbitrary number of references by Neyer (2010). An adaptive filter in the time domain was presented by Dalgaard et al. (2012) and compared to the timedomain optimal filter. They showed that both approaches led to similar results. All approaches have demonstrated a useful capability to improve the S/N ratio of surface-NMR data.However, a comparison that allows for determining the performance, i.e., pro et contra of each approach concerning different measurement conditions, is missing. Thus, to compare the time-domain with the frequency-domain approach, we developed a frequency-domain code beside the existing time-domain code and distinguished two strategies within the frequency domain. We present and describe in detail noise cancellation