Properties of the random noise source, which gives rise to inherent statistical fluctuations in nuclear reactors, have been studied under the assumption that the macrostochastic variables characterizing the state of the nuclear system follow the Markoffian random process. It has been found that the fundamental assumption leads to unified interpretation of phenomenological statements used repeatedly in the previous reactor-noise theory. They are: 1) the Langevin technique is to be applied; 2) the noise source is assumed to be white; 3) the Schottky formula is to be applied to determine the noise spectral density. Furthermore, the importance of the so-called Nyquist theorem is pointed out for establishing the Langevin method. The theorem shows that a generalized Einstein relation holds between the spectral density of the white-noise source and the linear constant operator describing the probable or expected kinetic behavior of nuclear systems. With the use of the relation, the noise spectral density has been classified into the binary and the single component. The latter comes from the fact that various nuclear reactions are of Poissonian nature, and produce the direct correlation term in the neutron field. The term is eliminated in the cross correlation function of the outputs of two detectors. The binary noise component, which comes from the branching processes and contributes to the count-rate fluctuations both for the one- and two-detector system of measurements, contains, however, the covariance of fluctuations of macrostochastic variables as unknowns. The complete determination of the noise source is accomplished with the use of the Smoluchowski consistency condition. The result offers a generalized Schottky formula. As an application, the space- and energy-dependent neutronic noise theory is treated in detail. Delayed neutrons are included from the outset. Applicability of the present theory to a slightly nonlinear system is suggested.