A first generation smoking machine capable of reading and replicating detailed puffing behavior from recorded smoking topography data is presented. Unlike standard smoking machines, which model human puffing behavior as a steady periodic waveform with a fixed puff frequency, volume, and duration, this novel machine generates a mainstream smoke aerosol by automatically "playing-back" puff topography recordings. Because combustion chemistry is highly non-linear, representing real smoking behavior with a smoothed periodic waveform may result in a tobacco smoke aerosol with a significantly different chemical composition and physical properties than that generated by a smoker. The machine presented here utilizes a rapid closed-loop control algorithm coded in Labview to generate smoke aerosols for toxicological assessment and inhalation studies. To illustrate its use, dry particulate matter and carbon monoxide yields generated using the playback and equivalent periodic puffing regimens are compared for a single smoking session by a 26-year-old male narghile water-pipe smoker. It was found that the periodic puffing regimen yielded 20% less carbon monoxide (CO) than the played-back smoking session, indicating that steady periodic smoking regimens, which are widely used in tobacco smoke research, may not produce realistic smoke aerosols.