We present a recently developed method of potential analysis of time series data, which comprises (1) derivation of the number of distinct global states of a system from time series data, and (2) derivation of the potential coefficients describing the location and stability of these states, using the unscented Kalman filter (UKF). We test the method on artificial data and then apply it to climate records spanning progressively shorter time periods from 5.3 Myr ago to the recent observational record. We detect various changes in the number and stability of states in the climate system. The onset of Northern Hemisphere glaciation roughly 3 Myr BP is detected as the appearance of a second climate state. During the last ice age in Greenland, there is a bifurcation representing the loss of stability of the warm interstadial state, followed by the total loss of this state around 25 kyr BP. The Holocene is generally characterized by a single stable climate state, especially at large scales. However, in the historical record, at the regional scale, the European monthly temperature anomaly temporarily exhibits a second, highly degenerate (unstable) state during the latter half of the eighteenth century. At the global scale, temperature is currently undergoing a forced movement of a single stable state rather than a bifurcation. The method can be applied to a wide range of geophysical systems with time series of sufficient length and temporal resolution, to look for bifurcations and their precursors.