In this paper, two cost functions are suggested for the parameter estimation of chaotic dynamical systems based on the least squares and state space. These cost functions, named ordinary least squares and total least squares,are orders of magnitude faster and more efficient than the previously suggested cost functions in the literature. They handle chaotic systems exceptionally well, contain no internal system model, work with noisy data, are easy toimplement and fast to optimise. The proposed cost functions are tested on three dynamical systems: the logistic map, the discrete nonlinear map that models neurodegenerative diseases and the Hénon map. The experimental results are then compared with other cost functions.