The distributed lag effect of ambient particulate air pollution that can be attributed to all cause mortality in Kathmandu valley, Nepal is estimated through generalized linear model (GLM) and generalized additive model (GAM) with autoregressive count dependent variable. Models are based upon daily time series data on mortality collected from the leading hospitals and exposure collected from the 6 six strategically dispersed fixed stations within the valley. The distributed lag effect is estimated by assigning appropriate weights governed by a mathematical model in which weights increased initially and decreased later forming a long tail. A comparative assessment revealed that autoregressive semiparametric GAM is a better fit compared to autoregressive GLM. Model fitting with autoregressive semi-parametric GAM showed that a 10 μg m-3 rise in PM10 is associated with 2.57 % increase in all cause mortality accounted for 20 days lag effect which is about 2.3 times higher than observed for one day lag and demonstrates the existence of extended lag effect of ambient PM10 on all cause deaths. The confounding variables included in the model were parametric effects of seasonal differences measured by Fourier series terms, lag effect of mortality, and nonparametric effect of temperature represented by loess smoothing. The lag effects of ambient PM10 remained constant beyond 20 days.