Tossing, throwing, or flicking objects in a user interface or virtual environment can be used as a faster, lower-precision alternative to traditional pointing, however there is currently no predictive model of user performance with tossing. We report experimental measurements of performance in a 1D tossing task from which a predictive model is derived. We consider a simplified form of tossing where a virtual object on a horizontal surface is accelerated and released, and then decelerates under friction, coming to rest at some final po- sition. The distance traveled after release is determined by the release velocity as well as by the friction model used. To abstract away the details of the friction model, our ex- periment measures the ability of users to accelerate and re- lease a virtual object in 1D (using a mouse) with a given tar- get velocity, with target velocities varying from 6.25 cm/s to 1 m/s. Results indicate that there is a linear relationship be- tween the target release velocity and the standard deviation of the release velocity achieved by the user. We also propose an automatic release technique (instead of requiring the user to manually release using a mouse button) that significantly improves precision. The model derived from our experiment predicts that a user should be able to toss at three different target speeds (effectively tossing toward target locations at three different distances) with an error rate under 4%. We also predict that having four or more targets in the same di- rection would cause the error rate to rise above 10%. Design implications for integrating tossing into graphical user inter- faces are discussed.