At the aggregate-data level, time-series analysis of annual data for the period 1950-90 gives little support for a relation between alcohol consumption and suicide in Switzerland. Input series in the analysis were per capita consumption and rates of unemployment and divorce. Output series were gender-specific and age-adjusted suicide rates. Different time-series techniques were used to assure stationarity and to avoid spurious correlation because of common trends. The Box-Jenkins approach on differenced data was compared with simultaneous equation modelling, where time trends were modeled by means of dummy variables. The effects were weak and unstable for different methods. The advantage of simultaneous equation models is discussed. Furthermore, a single "harmonic" long-term wave explained more of the variance in the suicide rates than any input variable examined. Its implication for a "natural suicide rate" is considered.