Random samples of hilsa were collected directly from fishers or from local markets throughout Bangladesh. Fish from markets were only taken where the site of capture was known. No juvenile fish were used in the study and all Bangladesh fish were between 1 and 2 years old. For the allozyme study, Chandpur was sampled three times. Morphometric samples were usually a subset of 30 fish taken from the allozyme samples. Once purchased, Bangladeshi fish were iced, transported to the laboratory and stored at −20?C until processed. Samples from India, Indonesia, Kuwait and Myanmar were obtained as whole frozen fish and processed in Australia. Sample sizes, collection dates and locations are listed. In the laboratory, morphometric and meristic traits were measured or counted prior to dissection of about 1 g of white muscle, heart, liver and eye (whole) tissues. Only liver and muscle were collected for most samples after initial screening of tissues revealed variable loci could be scored without eye and heart muscle. About 1 g of dissected tissue was placed into small, 1.5 ml plastic tubes and three to five drops of homogenising buffer (0.1 M Tris, 0.1 M EDTA, 50 µM NADP+, pH 7.0) were added to the samples. The vials were transferred to liquid nitrogen until the electrophoretic laboratory was reached, where the samples were then kept at −20?C. Gels were made with 10% potato starch (StarchArt, USA) in a specific buffer solution. Before each gel run, tissues were partially thawed, then centrifuged at 10000 × g for a few minutes at 4?C. Supernatant was soaked onto paper wicks of about 2–3 mm width, allowing 45 samples per gel. The rest of the electrophoretic protocol and histochemical staining used in this study. Thirty-five specific enzymes were examined during the initial survey to identify allelic variation. Five buffer systems were used to obtain optimum resolution of enzyme stains. Ten enzymes had insufficient activity to be scored reliably. The remaining 25 enzymes yielded information on 42 loci. Of these 25 enzymes, G6PDH, PGDH, AH, AAT, FH and haemoglobin, proved either unscoreable or inconsistent in their activity. Five variable loci were successfully screened using both TC-1 (0.135 M Tris, 0.043 M citric acid monohydrate, pH 7) and TM (0.10 M Tris, 0.1 M maleic acid, 0.01 M disodium EDTA, 0.01 M MgCl, NaOH adjusted to pH 7.4 with NaOH) buffers to optimize allele resolution. The most common allele at each locus was assigned a relative mobility of 100 and the mobility of all other alleles was calculated relative to this allele. All genetic analyzes were undertaken with two software programs: “Genes in populations” version 2 program, and ‘GENEPOP’ v3.1. During early screening of the Goalando, Cox’s Bazar, Khulna and Kuwait samples, the medium fast Idh-l allele (relative mobil-ity 108 on TC-1) and the medium slow Pgm allele (relative mobility 64 on TC-1) were not distinguished and so they were pooled with the nearest allele for analyzes. Observed and expected genotypic proportions were tested for agreement with Hardy–Weinberg equilibrium using a log-likelihood ratio G-test. Population structure was tested using the index of fixation, Fst and G statistics calculated by “Genes in populations” were tested for significant deviations from zero and from expected values. The three Chandpur collections were intended to accumulate sample sizes of around 100 per location but since they covered more than 1 year, they also served as a test of temporal variation. Initially, data from all sites were compared to see if there was significant genetic heterogeneity between sites. Then a series of adjacent pairwise comparisons was conducted, where appropriate, with Sidak’s multiplicative inequality correction for P (0.05) used in “Genes in populations”. This is used when testing the total G-value for significance similar to a Bonferroni correction. A simple Bonferroni Probability correction was applied to all significant tests from the Hardy–Weinberg and Genic diversity tests in “GENEPOP”. Isolation by distance was tested using the ISOLDE option in GENEPOP and the closest point-to-point distance between all sample sites. To simplify the analysis, two sample sites in Bangladesh were grouped as Paikgachha (Paik-gachha and Khulna), Goalando (Goalando, Rajshahi and Gaibhanda). This information was graphed in Excel with the Mantel statistic (Fst) plotted against he natural log of the distances between sites. Ten morphological and two meristic characters were measured (±0.1 mm) and counted on at least 20 fish from each sample. These were standard length (SL), head length (HL), snout length (SnL), inter-orbit cross-section (IOS), maxilla length (ML), body cross-section (BXS), body depth (BD) (both at dorsal fin origin), pelvic fin–anus distance (PA), caudal peduncle height (CPH), caudal peduncle length (CPL), pectoral fin rays (PFR), post-pelvic scutes (PPS). These measurements were chosen to facilitate comparisons with previous studies of hilsa morphometrics in Bangladesh. Standardised morphometric measurements were weighted by the inverse of the sample size and analyzed by PCA in SAS. The mean component scores of each sample on the first two components were found to be homoscedastic and so were compared separately between sites and between sexes by two-way ANOVA. Samples collected in Bangladesh were analyzed separately and site groups with similar dispersion on the two components were grouped. These groups were then analyzed with the samples from elsewhere. Mean ± 95% confidence limits of each site group were plotted. Where the confidence limits overlapped, an ellipse was drawn to enclose all groups. The degree of similarity among samples in the overall analysis and the relative importance of each measurement for group separation were assessed by stepwise discriminant function analysis (DFA) with cross-validation. The standardized morphometric measurements of fish from Bangladesh were also plot-ted to see if there was any evidence of hilsa showing the ‘fat’ and ‘thin’ shapes identified.