Systematic observation of behavior is a technique widespread in development clinical psychology. However, the issues concerning a reliable coding are barely described in a clear and systematic way. This contribution proposes a review that, without neglecting the corrected psychometrics bases, organizes the traditional knowledge and on it built an up-to-date and critical picture of the state of art on the observer reliability. Along with the reference concepts (agreement, calibration, reliability, precision, stability, accuracy, etc.), it treats the main coefficients proposed in the literature - first of all the Cohen's "K" - and show their advantages and limitations.