In this thesis, analysis of observed chaotic data has been investigated. The purpose of analyzing time series is to make a classification between the signals observed from dynamical systems. The classifiers are the invariants related to the dynamics. The correlation dimension has been used as classifier which has been obtained after phase space reconstruction. Therefore, necessary methods to find the phase space parameters which are time delay and the embedding dimension have been offered. Since observed time series practically are contaminated by noise, the invariants of dynamical system can not be reached without noise reduction. The noise reduction has been performed by the new proposed singular value decomposition based rank estimation method.Another classification has been realized by analyzing time-frequency characteristics of the signals. The time-frequency distribution has been investigated by wavelet transform since it supplies flexible time-frequency window. Classification in wavelet domain has been performed by wavelet entropy which is expressed by the sum of relative wavelet energies specified in certain frequency bands. Another wavelet based classification has been done by using the wavelet ridges where the energy is relatively maximum in time-frequency domain. These new proposed analysis methods have been applied to electrical signals taken from healthy human brains and the results have been compared with other studies.