In this work, a new, physically based model for the low-frequency noise is investigated by statistical simulations. The proposed model is based only on superposition of generation-recombination centers, and can predict the frequency-, current- and area-dependence of the low-frequency noise, as well as the area-dependence of the variation in the noise level. Measurements on Bipolar Junction Transistors (BJTs) are found to be in excellent agreement with the simulated results. For devices with large emitter areas AE, the model predicts a spectral density SIn ~ 1/f. For devices with submicron AE, SIn strongly deviates from a 1/f behavior, and several generation-recombination centers dominate the spectrum. However, the average spectrum <SIn>, calculated from several BJTs with identical AE, has a frequency dependence ~ 1/f. The extracted areal trap density within the frequency range 1-104 Hz is nT = 3 × 109 cm -2. The simulations show that the condition for observing g-r noise in the spectrum, strongly depends on the number of traps NT, as well as the distribution of the corresponding energy level for the traps. The relative noise level is found to vary in a non-symmetrical way around < SIn>, especially for small AE. For AE < 0.1 μ m 2, the model predicts a relative variation in the noise level [Formula: see text] below <SIn>, and [Formula: see text] above <SIn>. For AE > 0.3 μ m 2, the variation is found to be [Formula: see text].