Experimental site To investigate the long-term fertilization effects on rice yields and soil properties, paddy soil was prepared at the Bangladesh Rice Research Institute (BRRI) (230 85.9` N and 90? 82.4`E, elevation 12m), Gazipur, Bangladesh in 1985. The annual rainfall is approximately 2000 mm, and about 80% of rainfall is concentrated during the wet season from mid-June to the end September. The soil of the experimental field is Chhiata clay loam, a member of the fine, Hyperthermic Vertic Endoaquept. It has the typical chemical properties of Bangladesh rice paddy soil with clay loam texture, pH 6.8, and 11.3 g kg-1 organic C content. Experimental design and Fertilization Chemical fertilizer treatment (NPK) and comparison treatments (NP, PK, NK, and non-fertilized control) were employed. They were arranged in a completely randomized block design with four replications. The unit plot size was 6m x 7m. Two types of rice cultivars were grown in the experimental field under fully irrigated conditions. The Dry Season (DS) rice variety, BR3, was transplanted in the first week of January and harvested in May. The Wet Season (WS) rice variety, BR11, was transplanted in the first week of August and harvested in the third week of November. Two to three rice seedlings (45–50 days old in dry season and 30–35 days old in wet season) were transplanted at 20cm x 20 cm spacing. At maturity, 16 hills from each plot were harvested manually. The grain and straw were separated to evaluate yield properties. After harvest, the crop residues were incorporated into the soil. For NPK, N-P-K was applied at the rates of 120-25-35 kg ha−1 , respectively, during dry season, and 80-25-35 kg ha−1 , respectively, during the wet season. For comparison, the control was not fertilized. The NP, PK and NK treatments have the same nutrient composition as NPK, except for the absence of K, N and P, respectively. Nitrogen was applied as urea in three equal splits: (1) at the final land preparation before rice transplanting, (2) the active tillering stage, and (3) one week before panicle initiation stage. The total P and K were applied as basal fertilizers before rice transplanting by using triple super phosphate and muriate of potash, respectively. Rice yield and Nutrient Efficiency Grain and straw yields were determined at physiological maturity from 5 m2 areas within each plot following the standard method [25]. Grain yield was recorded after reducing the moisture content to ca. 14% (wt wt-1), and straw weights were expressed on an oven-dry basis (650C). The apparent crop Recovery Efficiency of the applied nutrient (RE) was determined by: R = (Cf − Cc)× 100/C, where R = % of recovered N, P, and K, Cf = absorbed N, P, and K by the rice for NPK plot, Cc = absorbed N, P, and K for PK, NK, and NP plots, respectively, and C = total amount of N, P, and K applied to NPK plot [25, 5]. Physiological Efficiency of applied nutrient (PE) was determined by: P = (Bf-Bc)/ (Cf − Cc), where P = % of physiological N, P, and K, Bf = Rice yield for NPK plot, Bf = Rice yield for PK, NK, and NP plots, Cf = absorbed N, P, and K by the rice for NPK plot, Cc = absorbed N, P, and K for PK, NK, and NP plots, respectively, Agronomic Efficiency of applied nutrient (AE) was determined by: A= RE-PE, where A= % of agronomicl N, P, and K, RE = % of recovered N, P, and K, and PE = % of physiological N, P, and K, Chemical Analysis Soil samples were collected from the surface layer (0-15cm) before the test in 1985 and after harvest in the 23rd year after the installation, and air-dried. The sieved soils (<2 mm) were analyzed for pH (1:2.5 water extraction), and organic carbon content (Walkley and Black method;). Total nitrogen, available phosphorus [18] and exchangeable potassium by ammonium acetate extraction method [19]. Dried rice grain and straw were ground to pass through a 0.5- mm sieve for quantifying nutrient contents. Total N content was analyzed by a micro-Kjeldahl method, and a ternary solution (HNO3:H2SO4:HClO4, 10:1:4 v v-1 ) was used for the determination of total P and K contents of plants. Statistical Analysis Analysis of variance (ANOVA) was performed by using the procedure of GLM to determine the effects of treatment on yield. Means for rice yield and soil properties were compared by using Tukey’s HSD method. Linear regression analyses were done to determine trends (slopes) of rice yield over the years using SAS systems. The P values, t-statistics, and 95% confidence intervals on the slopes were used to test whether the observed changes were significantly different from zero.