The Wildlife picture index (WPI) is presented by O'Brien et al. (2010) as a reasonable approach for monitoring medium and large-sized birds and mammals in a variety of habitats, with emphasis on tropical areas. The description of WPI is exemplary in several respects. It includes clear-headed discussion of why this sort of biodiversity monitoring is needed, what characteristics a biodiversity indicator should have, and how such a monitoring program can be implemented. Special attention is devoted to the ‘how’ question, as O'Brien et al. (2010) consider the two key design issues, spatial sampling and detectability. By relying on the occupancy modeling framework (MacKenzie et al., 2006; Royle & Dorazio, 2008), programs that adopt WPI will become some of the only biodiversity monitoring programs to explicitly incorporate detection probability into inference procedures. O'Brien et al. (2010) also present explicit recommendations for data analysis, again distinguishing their proposal from the vast majority of existing monitoring programs that have been designed and implemented in the absence of a clear idea of how resulting data are to be analyzed and used to meet program objectives. The example analysis of data from Bukit Barisan Selatan National Park provides potential users of this approach with a good idea of what sorts of results may be expected from a WPI program (O'Brien et al., 2010). In summary, the treatment of the proposed WPI by O'Brien et al. (2010) is excellent and includes discussion of the important issues that should be considered by those developing and initiating new monitoring programs. O'Brien et al. (2010: p. 1) begin their description of WPI with the following statement designed to motivate interest in their monitoring approach: ‘worldwide, biodiversity is being lost at a rate comparable in magnitude only to a handful of cataclysmic mass extinction events in the earth's geological history …’ This is a compelling statement that should motivate a strong desire to implement actions designed to reduce the loss of biodiversity. Note that this motivation does not depend on a precise estimate of the rate of loss; recognition of the general magnitude of loss should be adequate to stimulate serious conservation action. I raise this issue because I anticipate that some individuals and organizations will view WPI as a metric suitable for incorporation into omnibus monitoring programs designed to simply provide information about loss of biodiversity worldwide, and I do not view such omnibus monitoring as the most efficient way to spend limited conservation funds and effort. Some advocates of omnibus monitoring claim that precise estimates of rates of loss of biodiversity are needed to convince the general public that a problem exists. I fundamentally disagree, believing that funds expended on advertising and public relations are much more likely to sway public opinion than reducing the width of a confidence interval on an estimate of rate of species loss. Other advocates of omnibus monitoring argue its importance for prioritizing species and ecosystems deserving of conservation action. Certainly prioritization can be useful, however, the coarse rankings that are currently used (e.g. threatened vs. endangered species) should be adequate for most purposes. It is not productive to devote much effort to deciding which of two critically endangered species actually has the higher probability of extinction. Monitoring is most useful as a component of larger programs of conservation or management (Yoccoz, Nichols & Boulinier, 2001; Nichols & Williams, 2006). Informed programs of management are characterized by a set of essential components: objectives, actions, models, monitoring and an algorithm for selecting the appropriate action. In such a conservation program monitoring serves four different roles (see Yoccoz et al., 2001; Nichols & Williams, 2006). (1) Monitoring frequently provides estimates of system state variables required for state-dependent decisions. For example, very different actions may be recommended if WPI is very large versus very small. (2) Monitoring is useful for assessing the degree to which conservation objectives are being met. For example, O'Brien et al. (2010) suggest the use of WPI for assessing protected area effectiveness in achieving biodiversity objectives. (3) In the case of recurrent decisions (actions are taken not just once, but are applied to a system periodically), monitoring provides a basis for learning about system responses to management actions (Walters, 1986; Williams, Nichols & Conroy, 2002; Williams, Szaro & Shapiro, 2007). For example, one hypothesis may assert a strong relationship between protection and biodiversity, whereas an alternative (no management effect) may predict little difference between changes in biodiversity on protected areas and similar areas with no protection. Predictions of the different models can be compared with estimates of the state variable(s) of interest, permitting a formal updating of our degrees of faith in the different models (e.g. increasing faith in model(s) that predict well). (4) Monitoring is finally useful in providing estimates of vital rates and other parameters used to update models, when appropriate. Consideration of these specific roles of monitoring as a component of a program of conservation, leads to very specific designs that are tailored to the larger program. Such tailoring insures efficient expenditure of monitoring effort and funding. Omnibus monitoring programs are frequently justified by claims that they provide information useful to management, and this is likely to be true to some extent. However, monitoring tailored to, and designed for, specific management programs will be far more useful to those programs and represent more efficient use of monitoring efforts than omnibus monitoring. Finally, although monitoring should be designed to serve these four central roles in informed conservation processes, such ancillary outcomes as influencing public opinion, assessing general status and prioritizing conservation actions are useful by-products of targeted monitoring. I very much appreciated the statistical development presented by O'Brien et al. (2010), as they seriously considered most of the important implementation and analysis issues associated with WPI. The following discussion is devoted to various topics that may be worthy of additional consideration when implementing WPI. The basic elements incorporated into WPI are site- and species-specific estimates of change in occupancy over time (O'Brien et al., 2010). Specifically, the rate of change in occupancy for each species for a particular site and year is estimated as the ratio of the occupancy estimate for that site and year to the occupancy estimate at that site for some initial or reference year. Values that can be attained by this metric (rate of change in occupancy) are dependent on the occupancy in the initial or reference year. If occupancy of one species in the reference year is 0.4, then a growth rate of up to 2.5 is possible, whereas a reference year occupancy for another species of 0.9 would never permit a growth rate larger than 1/0.9≈1.11. There is nothing wrong with this bounding of potential growth rate by initial occupancy, but an alternative growth rate metric that avoids this issue is the odds ratio for occupancy estimates in the focal and reference years (see discussion in Mackenzie et al., 2006: p. 200). O'Brien et al. (2010) did not really specify whether their occupancy growth rate estimates were to be obtained as ratios of single-season estimates (the ‘implicit dynamics model’ of MacKenzie et al., 2006: pp. 186–187; O'Brien et al., 2010 appear to use this approach) or from a dynamic process model that includes local extinction and colonization (the ‘explicit dynamics model’ of MacKenzie et al., 2006: pp. 187–189). The explicit dynamics approach merits consideration for at least two reasons. (1) For conservation we are typically interested in ‘knowledge with which to respond to underlying drivers of loss’ (O'Brien et al., 2010: p. 1). In most cases, such knowledge is more readily obtained by focusing on factors affecting losses (extinction) and gains (colonization) than on those affecting overall rate of change. (2) Single season models assume that all sample units that share the same habitat covariates have the same probability of being occupied. However, the explicit dynamics modeling approach permits additional heterogeneity in occupancy probability associated with occupancy status the previous time step. Units previously unoccupied are permitted to have different probabilities of being occupied than units that were previously occupied. Following Buckland et al. (2005), O'Brien et al. (2010: p. 4) propose use of generalized additive models (GAMs; Hastie & Tibshirani, 1990) as an approach for inference about trends in WPI. The WPI themselves are trends, so the GAM modeling is essentially modeling trends of trends. Because detection probability differences across species and sites are incorporated into the occupancy modeling (thus eliminating a need to try to standardize for variation in detection), I recommend consideration of modeling trend of occupancy itself. I also recommend considering the direct modeling of trends using explicit dynamics models. Single-species models (MacKenzie et al., 2006: pp. 200–201) provide a one-step modeling process that properly accounts for the variance–covariance structure of all of the parameters of the likelihood. The tools are now available to model changes in occupancy for all species simultaneously, obtaining estimates of WPI-type metrics directly, while sharing information across species as appropriate (Royle & Dorazio, 2008; Zipkin, DeWan & Royle, 2009).