The invention brings forward a drawing method for a trend curve of large dataset used for computer data statistics analysis. According to the method, sample interval is firstly determined,
time quantum is equally divided, and characteristic value is determined. Firstly, pretreatment is carried out to sampled value of the
time quantum, and some values are ignored; then three centroids K1, K2 and K3 are selected from the
time quantum, clustering is carried out on the kept values, each value is classified to the clustering represented by the nearest
centroid to the value; after clustering, the first value and the last value of the K2 and K3 clustering are taken as characteristic constants, screening is carried out to the kept values of K2 and K3 to obtain the characteristic value; lastly, the number of characteristic value is determined, and curve is drawn by utilizing a Bessel
curve approximation algorithm. According to the invention, minimum sample values are adopted to draw the most approximate curve, independent of querying all data at a time and reading all data into memory at a time. The method provided in the invention greatly saves
system overhead and realizes rapid drawing of large dataset.