This paper investigates the use of price intensities to estimate volatilities based on high-frequency data. We interpret the conditional probability for the occurence of a price event within a certain time horizon as a risk measure which allows us to obtain an estimator of the conditional volatility per time. this kind of volatility estimation solves the problem of an appropriate eggregation level by defining explicitly price events. To consider grouping caused by the nontrading period overnight we use a categorical duration model. This model allows us to take into account that durations which occur overnight can only be registered by a lower and an upper bound. The use of price durations based on different tick sizes make it possible to investigate volaility patterns depending on different aggregation levels. Seasonalities are taken into account by including regressors based on a flexible Fourier form based on intraday and time to maaturity seasonalities. Testing for serial correlation and controlling for unobservable heterogeneity permits us to check for misspecification on different aggregation levels. Empirical results are based on intraday transaction data of Bund Future trading at the LIFFE in London.