Location and objectives: The experiment was conducted at Bangladesh Agricultural University Poultry Farm, at Mymensingh. The objective was to evaluate the compound broiler feeds based on visual observations, chemical analyses and feeding trials with broilers. Selection of feed samples: Out of 40 feed mills engaged in the production and marketing of poultry feeds, six were selected randomly and considered as the experimental treatments. AF = Aftab Feed Limited, Koliachar, Bazitpur, Kishoregonj. QF = Quality Feed Limited, Shirirchala, Bagher Bazar, Gazipur. PF = Paragon Feed Limited, Baniarchala, Bhabanipur, Gazipur. NF = Nourish Feed Limited, Gusinga, Sreepur, Gazipur. FF = Fresh Feed Limited, Meghnaghat, Sonargaon, Narayangonj. SF = Sundarban Feed, LimitedAmbug, Zampur Sonargon, Narayangonj. Collection of feed samples: Feed samples were collected from the sales and display centers of the feed mills. At the beginning of the feeding trial, the required amounts of broiler starter feed were procured. The same procedure was followed in the procurement of grower and finisher diets. Evaluation of the quality of compound feeds: After the reception of the feeds, the following three types of tests were performed. Visual observation: Physical characteristics (color, particle size, texture, flavor, odor, taste, presence of foreign particles, mould, fungus, insects and pests) were observed carefully during the collection of the feeds. Chemical analysis: Proximate analysis of feeds was done according to the procedures described by AOAC (1990). Calcium and total phosphorus were determined by atomic absorption and spectrophotometry methods, respectively (FAO 1980). Amino acid profiles of feed samples were determined by HPLC method according to AOAC (1990). Bioassay: 390 Shaver Starbro day-old broiler chicks were obtained from a commercial hatchery and were randomly divided into 6 treatments representing the 6 feed mills. There were 13 birds in each of 5 replicate groups. The experimental house was divided into 30 pens of equal size (1.4 m. x 0 .9 m.) by using bamboo materials and wire netting. The floor space was 0.097 m2/bird. Fresh dried rice husk litter was spread on the floor of the pens at a depth of about 4 cm before placing the chicks in the pens. To maintain brooding temperature, each small pen was equipped with two 100-watt electric bulbs suspended 0.4 m above the litter. The broiler starter diet was provided for the first 21days, followed by the grower diet from 22 to 35 days, then the finisher diet up to marketing age. In all cases, drinking water and feeds were supplied to the birds ad-libitum throughout the experimental period. One feeder (0.60 m × 0.08 m × 0.05 m) and one round drinker with a capacity of 2.5 liters were provided in each pen. Fresh drinking water was supplied to the birds three times a day, at 6 am, 12 am and 6 pm. Drinkers were washed and dried up daily in the morning, and feeders were cleaned weekly. The birds were exposed to a lighting regime of 23 hours and 30 minutes. All the birds were vaccinated against Newcastle Disease, Infectious Bursal Disease and Infectious Bronchitis Disease. Dead birds were sent to the Department of Pathology, Bangladesh Agricultural University, Bangladesh, for post-mortem examinations. At the end of the experiment, 4 broilers (2 males and 2 females) from each treatment with live weight close to the average of the pens were selected and slaughtered for meat yield characteristics. During the experiment, temperatures and relative humidity were recorded daily at 6-hour intervals. Strict sanitary measures were followed during the whole experimental period. Weekly body weight gain, feed consumption, feed conversion ratio were determined. Mortality was recorded. Statistical analysis: Descriptive analysis was performed using mean and standard deviation for each outcome variable. Differences between treatments were evaluated using analysis of variance (Proc GLM in SAS version 8.2). When treatments were significantly different, Duncan's Multiple Range Test was used to separate treatment means (Steel and Torrie 1981). Assumptions for the analysis were evaluated using residuals plots and the Shapiro-Wilks test for normality. A 5% significance level was used.