diff --git a/NEWS.md b/NEWS.md
index f74eed63a8..2a9731c76e 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -96,6 +96,8 @@
 
 21. GForce grouping is again faster when running single-threaded, e.g. after `setDTthreads(1)`, [#3330](https://github.com/Rdatatable/data.table/issues/3330). The batching used to parallelize `gsum()` and `gmean()` is skipped when only one thread is available. Also, `mean()` of an `integer` or `logical` column by group no longer coerces the whole column to `double` first, which made it slower than in v1.12.0 regardless of the number of threads. Thanks @jangorecki for the benchmarks, @mattdowle for the report, and @ben-schwen for the fix.
 
+22. `fread(text=, yaml=TRUE)` and `fread(input=, yaml=TRUE)` now work when the data is a single character string containing newlines, rather than failing with `cannot open the connection`, [#7896](https://github.com/Rdatatable/data.table/issues/7896). Thanks to @Noskario for the report and @AnKiT-GaRG2 for the fix.
+
 ### Notes
 
 1. {data.table} now depends on R 3.5.0 (2018).
diff --git a/R/fread.R b/R/fread.R
index 4aa6a01ed0..3303e6d7fd 100644
--- a/R/fread.R
+++ b/R/fread.R
@@ -276,7 +276,7 @@ yaml=FALSE, tmpdir=tempdir(), tz="UTC")
     call_args = names(match.call())
     if (is.character(skip))
       warningf("Combining a search string as 'skip' and reading a YAML header may not work as expected -- currently, reading will proceed to search for 'skip' from the beginning of the file, NOT from the end of the metadata; please file an issue on GitHub if you'd like to see more intuitive behavior supported.")
-    yaml_res = .read_yaml_header(input, skip, verbose)
+    yaml_res = .read_yaml_header(input, skip, verbose, input_is_data = !identical(input, file))  # file is NULL when input is the data itself, #7896
     yaml_header = yaml_res$yaml_header
     n_read = yaml_res$n_read
     yaml_names = names(yaml_header)
@@ -450,10 +450,11 @@ yaml=FALSE, tmpdir=tempdir(), tz="UTC")
 }
 
 # nocov start. Covered only in other.Rraw
-.read_yaml_header = function(f, skip, verbose) {
+.read_yaml_header = function(f, skip, verbose, input_is_data=FALSE) {
   # create connection to stream header lines from file:
   #   https://stackoverflow.com/questions/9871307
-  con = base::file(f, 'r')
+  # f can also be the data itself (text= or input= with \n or \r). Decided by the caller rather than by looking for \n in f, since a file name may contain a newline, #7896
+  con = if (input_is_data) textConnection(f) else base::file(f, 'r')
   # NB: close() won't be right if 'f' itself can be a file connection (#561)
   on.exit(close(con))
 
diff --git a/inst/tests/other.Rraw b/inst/tests/other.Rraw
index 928bc8e63e..760cf02de7 100644
--- a/inst/tests/other.Rraw
+++ b/inst/tests/other.Rraw
@@ -395,6 +395,11 @@ if (loaded[["yaml"]]) {  # csvy; #1701. Was 2032-2033 in tests.Rraw, #5516
   test(16.22, fread(f, skip = 'var1,', yaml = TRUE),
        DT_yaml, warning = 'Combining a search.*YAML.*')
 
+  # yaml=TRUE with text= a single string containing newlines, #7896
+  txt = paste(readLines(f), collapse="\n")
+  test(16.23, fread(text=txt, yaml=TRUE), DT_yaml)
+  test(16.24, fread(txt, yaml=TRUE), DT_yaml)
+
   # fwrite csvy: #3534
   tmp = tempfile()
   DT = data.table(a = 1:5, b = c(pi, 1:4), c = letters[1:5])