The exchanges in the New Generation Optical Access Network (NG-PON), are mainly based on the bandwidth request by the Optical Network Units (ONU) and its granting by the Optical Line Termination (OLT). This process (request-grant) not being synchronized, will create in the ONU buffer a long delay and the loss of sensitive packets. Many authors, from signaling to prediction, have proposed models to control these parameters, however only those based on prediction reduce the delay in the queues but also waste bandwidth by over- or under-allocating. Our goal is to propose a model that accurately allocates bandwidth to ONU while minimizing their queue delay. As such, we propose a Statistical Queue Estimation model based on ONU Buffer Inputs and Outputs (ESTEST). In its execution ESTEST memorizes the ONU bandwidth requirements and the bandwidth used to transmit their data in the ith upstream frames. With these two parameters (requirements and bandwidth used), our algorithm computes the traffic intensity (𝜌) and predicts the queue size of the ONU. On 100 observations we compute the mean and standard deviation of the 𝜌th values and with the reduced centered normal distribution, find the confidence interval of size estimation.