—The forecasting of Particulate Matter (PM10) is crucial as the information can be used by local authority in informing community regarding the level air quality at specific location. The non-linearity of PM10 in atmosphere after it was subjected by several meteorological parameters should be treated with powerful statistical models which can provide high accuracy in forecasting the PM10 concentration for instance Neural Network (NN) model. Thus, the aim of this study is establishment of NN model using Levenberg-Marquardt training algorithm with meteorological parameters as predictors. Daily observations of PM10, wind speed, relative humidity, ambient temperature, rainfall, and atmospheric pressure in Kuala Terengganu, Malaysia from January 2009 to December 2014 were selected for predicting PM10 concentration level. Principal Component Analysis (PCA) was applied prior the establishment of NN model with the aim of reducing multi-collinearity among predictors. The three principal components (PC-1, PC-2, PC-3) as the result of PCA was used as the input for the NN model. The NN model with 14 hidden neurons was found as the best model having MSE of 0.00164 and R values of 0.80435(Training stage), 0.85735(Validation stage), and 0.8135(Testing stage). Overall the model performance was achieved as high as 81.1% for PM10 forecasting in Kuala Terengganu.