The objective of this thesis is to investigate experimentally and numerically the non-homogenous distribution of nanoparticles in the nanofluid by natural convection heat transfer inside a square enclosure. There are some phenomenon such as fluid motion, Brownian effect, thermophoresis effect and gravity which cause nanoparticles movements throughout the nanofluid. An experimental set-up was built to carefully take micro-litter samples of nanofluid from a 8×8×18(cm 3 ) square cavity. The nanofluid was prepared using 20nm-gamma type Al 2 O 3 nanoparticles dispersed in deionized distilled water. The experiments were done for three Rayleigh numbers 0.992×10 7 , 0.51×10 8 and 1.53×10 8 and the particle loading of 1% was constant during the experiments. The contours of nanoparticle fraction shows that the assumption of non-homogeneity is correct and the regions of high and low nanoparticle volume concentration are distinguishable. The fluid motion is the most important factor that may come in to play especially at higher Rayleigh number and causes the maximum differences between measured nanoparticle concentration. The result showed that the differences between maximum and minimum volume fraction in the whole cavity increases from 18% at Ra=0.992×10 7 to 28.76% at Ra=1.51×10 8 . It was observed that the average nanoparticle volume fraction along cold wall is 3.10% greater than that along hot wall for minimum Ra number however at higher Ra numbers this difference would weaken. To check precision of replicates and ensure accuracy, the data were analyzed using analysis of variance method (ANOVA method). In addition to experimental part, a numerical study has been conducted for two different geometry: A square cavity and a cylindrical enclosure. Numerical methods was used for the computational analysis of natural convection heat transfer from fire hot tube/plate to the cold tube/plate inside the water-Al 2 O 3 nanofluid in enclosure/cavity using alumina nanoparticles. The mathematical model is based on two-dimensional continuity, momentum, energy and volume fraction equations, which are solved numerically. The simulation was done for different value of particle loading 1%, 2% and 5% at Rayleigh numbers 10 3 , 10 4 ,10 5 ,10 6 , 10 7 and 10 8 . The results show that nanoparticles enhance the heat transfer by increasing the volume concentrations of particles in the case of enclosure at low Ra numbers while it diminishes at high Ra numbers in the case if cavity. It was observed that the maximum augmentation of the average Nusselt number is about 30.1% at the Ra=10 3 and 5% volume fraction. Although the average Nusselt number rises by increasing the Rayleigh number, the ratio of heat transfer using nanofluid to that by pure fluid decreases. Using 5% volume fraction of alumina nanoparticles at Rayleigh number of 10 3 increases the heat transfer to cold tube by about 23% compared to the pure water. The effect of nanolayer formation around particles was considered in thermal conductivity model which shows a maximum of 4.8% increase of the Nusselt number at the Ra=10 3 and 5% of particle loading. To verify the solution results, comparisons with previously published work on the basis of special cases are performed. The value of non homogeneity was 11.81% and 18% at the Ra=10 7 from numerical solution and the experiment respectively. Th maximum non-uniformity was observed at the maximum Ra number of 1.5×10 8 which was 28.76% from experiments and 24.30% for numerical solution. The present work reveals that there are still some issues for flow and thermal mechanisms in nanofluids, particularly regarding the nanoparticles distribution for future works. Keywords: Natural Convection, Non-homogeneous Nanofluid, Brownian Motion, Nanolayer Formation, Thermophoresis Effect, Heat Transfer.