Method and apparatus for the treatment of cardiac failure, Cheyne Stokes breathing or central sleep apnea are disclosed. A subject is provided with ventilatory support. Respiratory airflow is determined. From the respiratory airflow are derived a measure of instantaneous ventilation (for example half the absolute value of the respiratory airflow) and a measure of longterm average ventilation. A target ventilation is taken as 95% of the longterm average ventilation. The instantaneous ventilation is fed as the input signal to a clipped integral controller, with the target ventilation as the reference signal. The output of the controller determines the degree of ventilatory support. A third measure of ventilation, for example instantaneous ventilation low pass filtered with a time constant of 5 seconds, is calculated. Ventilatory support is in phase with the subject's respiratory airflow to the fuzzy extent that this ventilation is above target, and at a preset rate conversely.