Throughout earths history, the climate have always naturally varied due to both external forcings and internal mechanisms. With the outlook on future climate change, it is important to understand the climate system and this includes its natural variability. Depending on the driving mechanisms, the variability happens over different timescales. Among these is centennial variability, which still have unanswered questions. Studying centennial variability requires climate timeseries that are both long enough to encompass several cycles and detailed enough to resolve the variability signals. New data from long transient Holocene model simulations and recent efforts to compile large proxy databases now presents an opportunity to study centennial variability with better data foundation than earlier. To do so, this master thesis studies centennial variability in transient Holocene simulations from 9 models and 122 proxy records with a spectral analysis with the aim of finding the general signals related to cycle length, geographic dependencies and discuss the implications for the ongoing scientific discussion on the potential driver(s). The spectral analysis of the proxy records finds centennial variability significant from red noise in the majority of the records, with the highest concentration of cycle lengths around 120- 130 years, an average at 240-300 years, depending on the number of cycles included in the calculation, and no clear indication of it being dependent on which climate variable, although air temperature is the biggest group and influence the results the most. The analysis of the model global mean temperature (GMT) also finds centennial variability in all simulations with the highest concentration of cycle lengths around 120-150 years and an average just slightly above that. A good agreement between model and proxy data is thereby indicated, although the spread is slightly larger for the proxy data, but this is also a more diverse collection of data than the models. There is also a good agreement in the lack of latitudinal dependencies, where centennial variability is found at all latitudes of the model data (6 bands with a combined global coverage is analysed) and no clear differentiation is found between the proxy records at different latitudes. However, all the model data have most spectral density distributed over the 90N to 60N latitude band, which indicates either a particular variability sensitivity or potential driving mechanisms in this region. Four of the models also have differentiated/single forcings simulations and the spectral analysis of the GMT in all of these also reveals significant centennial variability with cycle lengths between 100-200 years. The simulations where only orbital forcing also show this and so the different forcings seem to induce some variability to the system, but none can be said to be the main driver based on the spectral analysis. This also includes solar irradiance, which long have been hypothesised to drive centennial variability, as all the simulations without this forcing, which includes some of the full forcing simulations as well, all have significant centennial variability. The results instead indicate that centennial variability is internally driven and that the Arctic is a region of interest for this aspect. The conclusions is not without uncertainties, as both proxies and model simulations have uncertainties, but when analysing with a largely uniform approach on a large data collection, there is good evidence of centennial variability with cycle lengths around 100-200 years across the entire range of available Holocene paleoclimate data