Climatic variations happen at all time scales and since the origins of these variations are usually of very complex nature, climatic signals are indeed chaotic data. The identification of the cycles induced by the natural climatic variability is therefore a knotty problem, yet the knowing of these cycles is crucial to better understand and explain the climate (with interests for weather forecasting and climate change projections). Due to the non-stationary nature of the climatic time series, the simplest Fourier-based methods are inefficient for such applications (see e.g. Titchmarsh (1948)). This maybe explains why so few systematic spectral studies have been performed on the numerous datasets allowing to describe some aspects of the climate variability (e.g. climatic indices, temperature data). However, some recent studies (e.g. Matyasovszky (2009); Paluš & Novotná (2006)) show the existence of multi-year cycles in some specific climatic data. This shows that the emergence of new tools issued from signal analysis allows to extract sharper information from time series. Here, we use a wavelet-based methodology to detect cycles in air-surface temperatures obtained from worldwide weather stations, NCEP/NCAR reanalysis data, climatic indices and some paleoclimatic data. This technique reveals the existence of universal rhythms associated with the periods of 30 and 43 months. However, these cycles do not affect the temperature of the globe uniformly. The regions under the influence of the AO/NAO indices are influenced by a 30 months period cycle, while the areas related to the ENSO index are affected by a 43 months period cycle; as expected, the corresponding indices display the same cycle. We next show that the observed periods are statistically relevant. Finally, we consider some mechanisms that could induce such cycles. This chapter is based on the results obtained in Mabille & Nicolay (2009); Nicolay et al. (2009; 2010).