Dr. Philip Loyd from South-Africa has published in Energy & Environment a short but easily understandable paper trying to ask this question (link to abstract). He took several reconstructed temperature series reaching back up to over 8000 years, i.e. covering the Holocene period. An example shown in the figure below is the Gips-2 reconstruction from a Greenland ice core, where temperatures have been deduced from an analysis of the Ar (Argon) and N2 (Nitrogen) isotopes.
First he de-trended the data by subtracting the best polynomial fit (which was a linear one in this case); than he calculated the differences from temperatures 100 years apart, and finally calculated the statistical standard deviation of this ensemble. The following figure gives the result for the 4 series that he used:
Taking these 4 results and computing the average gives 0.98 (rounded to 2 decimals) with an uncertainty of 0.27.
So the historical data show that temperatures fluctuate naturally by ~0.98 °C during one century; as the Hadcrut3 series for global temperatures from 1900 to 1999 gives a warming of 0.7°C, this number is smaller than the natural centennial variability. The author concludes prudently “the signal of anthropogenic global warming may not yet have emerged from the natural background”; which means in very simple words that human caused centennial warming (by greenhouse gas emissions for instance), if it exists, does not exceed for the moment natural temperature variability.