For an IEEE secondary citation, first trace the original work and read the passage behind the claim. If you cannot obtain it, follow the guide required by your journal or institution. RMIT and Newcastle give different instructions for this situation. We checked both guides and followed a public review to its original conference paper on October 19, 2026. The trace below separates source identification from actually checking the result.
Start With What You Actually Read
An IEEE secondary citation arises when you use an original author's finding through another publication. If you read a review that describes a 2012 experiment, you have read the reviewers' account. The original experiment becomes directly checked evidence only after you obtain the paper and inspect the relevant passage.
Decide what your sentence needs to establish. A sentence about how a review interprets a field can cite that review. A sentence reporting an experiment's numerical result should lead you to the experiment. Reviews are legitimate sources for synthesis; their bibliographies also provide routes to earlier evidence.
Do not copy a review's reference number into your manuscript. A number belongs to the reference list in the document where it appears. In the example below, reference 1 inside the review becomes [2] in our illustrative two-entry list.
Follow a Real Review to the Original Paper
Begin with LeCun, Bengio, and Hinton's 2015 review Deep learning. Its Nature record identifies the article, while Hinton's university site provides a public full-text copy. On printed page 442, reference 1 identifies the 2012 ImageNet paper by Krizhevsky, Sutskever, and Hinton.
- Copy the original paper's title and authors from that reference. Keep the review's page and reference number in your notes.
- Search the exact title. Open the official NIPS 2012 proceedings record and compare its title, authors, and year.
- Open the linked original paper PDF. Read section 6 and Table 2 for the ILSVRC-2012 comparison.
The table gives a 15.3% top-5 test error for the seven-CNN ensemble row, compared with 26.2% for the competing entry. Its notes explain the pretraining condition. Those details support a narrower sentence than an unspecified claim that one neural network cut image-recognition errors in half.
Krizhevsky et al. report a 15.3% top-5 test error for their seven-model ILSVRC-2012 ensemble [2].
This is a paraphrase after checking the original table. Keep the dataset, metric, and model configuration with the number. Reading an abstract or finding a citation export would establish less than inspecting the relevant result.
Check the Record Before Copying the Reference
The trace also reveals a bibliographic discrepancy. The review's reference 1 prints pages 1090-1098. The conference's own metadata gives first page 1097 and last page 1105. The example below uses that proceedings record. Copying the review's bibliography would carry the different range into your paper.
[1] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436-444, May 2015, doi: 10.1038/nature14539.
[2] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet classification with deep convolutional neural networks,” in Proc. Adv. Neural Inf. Process. Syst., vol. 25, 2012, pp. 1097-1105.
These entries illustrate a manuscript that cites the review first and the original paper second. Your numbering follows your own order of first citation. An original paper does not need the number assigned to it by the review.
For another trace, use the Google Scholar to IEEE guide to find candidate records. If a reference includes a DOI, resolve it and compare the landing page with the citation. The DOI to IEEE guide covers that identity check. A successful lookup still leaves the passage to read.
Record Three Different Evidence States
Keep a short source note beside each inherited claim. Record the work you read, the passage you inspected, and what remains unresolved. This prevents a downloaded citation file from being mistaken for a paper you have read.
| State | Evidence note | Writing decision |
|---|---|---|
| Original passage checked | Review p. 442, ref. 1 traced to NIPS 2012; original section 6 and Table 2 inspected | Cite the original for the checked result; retain the review for its own interpretation |
| Only the review read | Review passage inspected; original identified but its relevant text unread | Attribute the review's account; continue tracing before claiming direct verification |
| Original unavailable to you | Record the catalogue, repository, or library routes tried, the date, and the access limitation | Apply your required guide's rule for indirect citation or omit the unsupported original-source claim |
The first row describes our completed trace. The other rows are alternative reader situations, not claims that this publicly accessible paper was unavailable. A paywall at one publisher link establishes an access problem at that link; an author's university copy may still be available.
Format the Source After Checking the Evidence
Once you have identified the source you can legitimately cite, use the IEEE Citation Generator to format its supported fields. Compare authors and publication details with the source record, then match the generated number to your manuscript. The tool cannot establish that you read a paper or that a passage supports your claim.
A direct quotation requires checking the exact words in the version you cite. If you have only the review's wording, do not place it in quotation marks and attribute it to the original authors as though you inspected their text. Attribute the words to the text you could inspect.
Sources and Example Notes
Checked October 19, 2026. The university guides linked above establish the differing secondary-referencing policies. The public review, original paper, and official proceedings metadata establish the trace and page-range discrepancy. Example sentences are paraphrases, and the two-entry reference list is illustrative.