After many years spent by paleomagnetists studying the directional behavior of the Earth's magnetic field at all possible timescales, detailed measurements of field intensity are now needed to document the variations of the entire vector and to analyze the time evolution of the field components. A significant step has been achieved by combining intensity records derived from archeological materials and from lava flows in order to extract the global field changes over the past 12 kyr. A second significant step was due to the emergence of coherent records of relative paleointensity using the remanent magnetization of sediments to retrace the evolution of the dipole field. A third step was the juxtaposition of these signals with those derived from cosmogenic isotopes. Contemporaneous with the acquisition of records, new techniques have been developed to constrain the geomagnetic origin of the signals. Much activity has also been devoted to improving the quality of determinations of absolute paleointensity from volcanic rocks with new materials, proper selection of samples, and investigations of complex changes in magnetization during laboratory experiments. Altogether these developments brought us from a situation where the field changes were restricted to the past 40 kyr to the emergence of a coherent picture of the changes in the geomagnetic dipole moment for at least the past 1 Myr. On longer timescales the field variability and its average behavior is relatively well documented for the past 400 Myr. Section 3 gives a summary of most methods and techniques that are presently used to track the field intensity changes in the past. In each case, current limits and potential promises are discussed. The section 4 describes the field variations measured so far over various timescales covered by the archeomagnetic and the paleomagnetic records. Preference has always been given to composite records and databases in order to extract and discuss major and global geomagnetic features. Special attention has been devoted to discussing the degree of confidence to be put in the data by considering the integration of multiple data sets involving different techniques and/or materials.