Probabilistic approaches allow designing very efficient data structures and algorithms aimed at computing the number of flows within a given observation window. The practical applications are many, ranging from security to network monitoring and control. We focus our investigation on approaches tailored for sliding windows, that enable continous-time measurements independently from the observation window. In particular, we show how to extend standard approaches, such as Probabilistic Counting with Stochastic Averaging (PCSA), to count over an observation window. The main idea is to modify the data structure to store a compact representation of the timestamp in the registers and to modify coherently the related algorithms. We propose a timestamp-augmented version of PCSA, denoted as TS-PCSA, and compare it with state-of-the-art solutions based on Hyper-LogLog (HLL) counters that evaluate the cardinality over a sliding window, but without storing the timestamps. We will show that TS-PCSA with a limited memory footprint is achieving a different tradeoff between memory and accuracy with respect to HLL-based solutions.