We present a simple stochastic mechanism which generates pulse trains exhibiting a power law distribution of the pulse intervals and a 1/f α power spectrum over several decades at low frequencies with α close to one. The essential ingredient of our model is a fluctuating threshold which performs a Brownian motion. Whenever an increasing potential V (t) hits the threshold, V (t) is reset to the origin and a pulse is emitted. We show that if V (t) increases linearly in time, the pulse intervals can be approximated by a random walk with multiplicative noise. Our model agrees with recent experiments in neurobiology and explains the high interpulse interval variability and the occurrence of 1/f α noise observed in cortical neurons and earthquake data. [5,[7][8][9][10]] resemble a pulse train consisting of individual, largely identical events which occur at discrete times. This is especially true for spike trains of single nerve cells for which 1/f α noise has been observed in various brain structures [11][12][13][14][15]. The reported exponents, which depend both on the presence or absence of a sensory stimulus [11,12] and on the state of the animal (REM sleep vs awake state) [13][14][15], vary from 0.68 to 1.38. The power-law behavior for spike train power spectra lies within the range 0.01 to 10 Hz, extending typically over 2 decades. In almost all cases the upper limit of the observed time over which fractal correlations exist is imposed by the duration of the recording.In this Letter, we propose a simple mechanism for generating pulse trains with 1/f α behavior in systems with a threshold-controlled dynamics like, e.g., neurons and earthquake faults. Our model is based on an integrateand-fire (IaF) mechanism and consists of a single unit characterized by two variables (see Fig. 1): The voltage V (t) and the threshold C(t). Initially, the voltage is below the threshold. Then, the voltage increases monotonically in time -in the simplest case just linearly -and the threshold evolves according to a Brownian motion with diffusion constant D within reflecting boundaries V 0 < C l < C(t) < C u . As soon as V (t) has reached the threshold, the voltage is reset to V 0 and a pulse of unit height is emitted. In this way, a pulse train is generated. Note that the threshold is not reset to its initial value. Such a model is often used to describe single neurons: V (t) is the membrane potential and the emitted pulse is the generated action potential. However, it is usually assumed that the threshold is constant in time. Recent investigation have shown that this is not true for cortical neurons in vivo [16] and in vitro [17]. Additionally, there is evidence that the spike trains of auditory neurons [18] and of neurons in the mesencephalic reticular formation [19] are not renewal, i.e., successive time intervals between spikes are correlated. These facts are incorporated in our model in the simplest possible way. Moreover, the effects of dead time or absolute refractoriness, which limits the rate at which a neuron can fire, are...